Skip to content

Latest commit

 

History

History
435 lines (383 loc) · 98.9 KB

CHANGELOG.md

File metadata and controls

435 lines (383 loc) · 98.9 KB

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Added

  • pkg/subenum: Config.Validate() reports the error Run would return for a config, such as an unsupported type, an invalid Exclude pattern or ErrNoWords, without starting a scan or sending a query (#378).
  • -simulate-zone accepts the lab scenarios built into every binary by name, so release, container and go install users run the labs without cloning the repository: subenum -simulate-zone lab1 lab.example works from any directory (lab1, lab1-first-scan and lab1-first-scan.zone all name it), and without -w the scan uses the labs' own wordlist (-w "" keeps the bundled 5,000-word list). A path, or a file that exists under the given name, is still read from disk, and an unknown name lists the built-in scenarios (#354).
  • pkg/subenum: Result.Sources names what proposed each result, as the CLI's sources field does: wordlist, and each Config.Sources entry that returned it, named by a Name() method when it has one (ct, wayback, axfr for the built-in sources) or as Sources[i] (#273).
  • -o can be given more than once, and a format prefix gives a file its own format: -o names.txt -o jsonl:results.jsonl writes bare names for the next tool and JSONL with the run-quality report in one run, where a pipeline used to scan twice or extract names with jq. A file without a prefix uses -format, as before. Every file gets the full current results, also under -diff; on a run stopped with state for -resume, only an -o file that is also the -diff file is left unchanged. The same path given twice is a usage error, and -tui takes the first file (#334).
  • The TUI's finished screen can save the results of a scan run without an Output file: s asks for a path (starting from the Output file, or subenum-<domain> with the Format's extension), and the file is written as the Output file would have been, in the format its extension names (.txt, .json, .jsonl, .csv, otherwise the form's Format). With a filter on, only the results it shows are saved. v toggles each result's records under it, during or after the scan, and a text file saved while they are shown carries them, as with -show-records. Results past the view's 10,000 are not kept and cannot be saved (#332).
  • subenum mcp offers a read-only diff tool that compares two result sets without sending a DNS query: previous and current each take a scan or lab_scan report (or its results array) or the contents of a results file in any format, and an optional types limits the record types compared. It returns added, missing and changed names with their records, as structuredContent too, compared as subenum diff -records does, so CDN address rotation behind an unchanged CNAME is not a change; missing names are marked unverified. Its input is bounded by the server's 16 MiB message limit, as explain_stats's is (#333).
  • subenum diff [-records] [-type A,AAAA,CNAME] [-format text|json|jsonl|csv] <previous> <current> compares two saved results files, in any format subenum writes, without sending a DNS query: + for names only in the second file, - for names only in the first, and with -records ~ for names whose records differ, compared as -diff-records does. In JSON, JSONL and CSV the change is added, missing or changed; a name is missing rather than removed because nothing was looked up to show it is gone. It exits 4 when the files differ, and is in the shell completions and the man page (#310).
  • -resume accepts the pacing flags -rate, -rate-per-resolver, -t, -timeout, -attempts, -auto-rate, -no-abort and -max-duration next to the state file, so a scan stopped because timeouts piled up can continue at a gentler rate instead of the one that caused them. They apply over the saved command line, a changed value of one is no longer a reason to refuse the resume, and the next state file records them. Flags that change the candidates or the wildcard check stay refused (#309).
  • Library: a KindError event has Fatal set when it ends the scan (the reliability guard's warning under NoAbort is the one error without it), and KindDone has Failed set when the scan failed. The CLI, the TUI, Scan and subenum mcp now read these instead of each working it out from the errors seen and the no-abort setting; exit codes, Scan errors and verdicts are unchanged (#308).
  • The TUI configuration form includes Sources and Permute fields for -sources and -permute. The passive sources named in Sources (ct, wayback) are asked while the wordlist loads and their names are scanned with the wordlist's; a source that fails is a warning on the scan screen, and none is asked with Simulate on. With Permute on, a permutation pass follows the scan on the same screen, and the finished screen's verdict and counters cover both passes. With either field set, the results file names what proposed each result (sources, and permutation for a permutation hit) and its summary has the sources and permutation entries the CLI's report has (#274).
  • Each result says what proposed it: a sources list in JSON and JSONL and a last sources column in CSV (joined with ;), holding wordlist, ct, wayback and axfr, every one that had the name, or permute alone for a -permute hit. A name found by -recursive below another is the wordlist's. The run-quality report counts the results each passive source had proposed (found in its sources entry) and a zone transfer's (axfr_found). Text output, the TUI's files and subenum mcp reports are unchanged (#273).
  • subenum mcp: scan and lab_scan accept an offset that skips that many candidates at the start of the word list (after dedupe), and a report whose max_queries left words untested carries next_offset, named in its reason. An agent can now work through the bundled 5,000-label list in several calls instead of testing the same first -max-queries names each time; every call stays within -max-queries, -max-duration and -max-calls (#313).
  • The TUI configuration form includes Auto Rate, Max Duration and Retry Inconclusive fields for -auto-rate, -max-duration and -retry-inconclusive. Max Duration accepts a Go duration such as 20m or 1h30m; leave it blank for no time limit (#274).
  • subenum mcp serves subdomain enumeration to AI agents over the Model Context Protocol (stdio). Tools: lab_scan, which scans the compiled-in lab scenarios on a loopback DNS server; scan, offered only with -allow and refusing any domain outside it; and explain_stats, which explains a -stats report or -format jsonl output. The operator sets the resolver, rate, timeout and budgets (-dns-server, -rate, -timeout, -max-queries, -max-duration, which cannot be turned off). A call can lower max_queries, never raise it; unknown arguments are refused, and calls run one at a time. Each scan returns its results with a verdict (complete, degraded, unreliable or not_run), and sends progress notifications about once a second when the call carries a progressToken (#267). -sources ct,wayback lets scan calls seed candidates from passive sources through a sources argument; it is off by default, and the report counts each source's names or names its error (#268). scan and lab_scan accept retry_inconclusive, auto_rate, exclude and depth (up to 3), all within the operator's caps, so an agent can act on a verdict that is not complete; the report adds retried, recovered, rate_backoffs and excluded (#269). Their report is also declared as an outputSchema and sent as structuredContent, beside the same JSON as text (#270). -max-calls (default 100) caps the scan calls one session may make, and a client that stops reading replies ends the session, cancelling queued calls instead of scanning for nobody (#240). A request line over 16 MiB gets a JSON-RPC parse error instead of ending the session, a request reusing the ID of one still running is refused, and a scan cancelled before testing any name reports not_run rather than degraded (#241). The JSON-RPC layer is a small hand-written internal/mcp package, so there is no new dependency (#131). lab_scan's scenario argument describes what each lab teaches, taken from the opening comment of its zone file (#275).
  • -sources ct,wayback picks keyless passive sources whose names become candidates and -permute seeds, each still verified by a lookup. ct is crt.sh, as -ct, which stays as shorthand. wayback is new: the hostnames in the Wayback Machine's index of archived URLs, one HTTPS request per target. The report gains a sources entry per source (names, or the error it hit; ct_names and ct_error are kept), a failed source is a warning, the resume state keeps each source's names, simulation ignores -sources and lab mode refuses it. Sources share one interface in a new internal/passive package. pkg/subenum gains Config.Sources, the Source interface, CertificateTransparency(), WaybackMachine() and NoticeSource (#193).
  • -axfr tries a zone transfer (AXFR) of each target from its name servers before the scan: the NS records come from -dns-server, then each server (up to 8, two addresses each) is asked over TCP. A server that allows it is reported on stderr and in the run-quality report (axfr_allowed, axfr_names), and the zone's names become candidates and -permute seeds, each still verified by a lookup. Refusals, the correct configuration, stay quiet unless -v. The NS and address lookups and the transfers count in queries_sent; simulation ignores -axfr, lab mode refuses it, and a resumed scan reuses the transferred names. The test DNS server can answer AXFR (Reply.Transfer) (#85).
  • -axfr also sends one DNSSEC probe per target to an authoritative server: a query for a name that does not exist, with the DNSSEC OK bit set. A zone that proves the absence with plain NSEC can be walked to list every name, so it is reported as Zone walkable on stderr. The report's zone_denial says nsec, nsec3 or none (unsigned). subenum does not walk the zone. The test DNS server can sign negative answers (Reply.Denial) (#234).
  • pkg/subenum exports Diff(previous, current []Result, types []string) Changes: the names added and missing and the names whose records changed, compared the way -diff-records does (as sets over the given types, CNAME targets only when an alias is involved). It sends no DNS queries, so Removed lists names to look up again rather than names known to be gone. The CLI and the library share the comparison in a new internal/diff package (#197).
  • The TUI scan view can filter its results: / types a name filter, enter keeps it, esc clears it, and t toggles takeover candidates only. While a filter is on the view stops following new results and the hint line says showing N of M. The results file is unaffected (#221).
  • -auto-rate keeps a scan going when its resolver starts rate-limiting: when more than 20% of the last 200 lookups failed, the query rate is halved (from the rate reached so far when -rate is 0) with a rate notice, and raised again by a quarter at a time, up to where it started, once a window of lookups is healthy. Only after 6 back-offs, or at 5 queries per second, does the reliability guard abort. The report gains rate_backoffs and final_rate, and pkg/subenum gains Config.AutoRate, NoticeRate, Stats.RateBackoffs and Stats.FinalRate. Without the flag nothing changes (#220).
  • -type accepts NS, MX and TXT as well as A, AAAA and CNAME. MX values read <preference> <host>. With NS, a delegated subzone whose name server does not exist (NXDOMAIN) is marked dangling-ns, a classic takeover that CNAME-only scans miss, and -sarif reports it under a subenum/dangling-ns rule. A recursive resolver usually answers SERVFAIL for a subzone whose every name server is gone, so that case shows up as inconclusive rather than marked. Each new type costs one query per name under -rate. The lab scenario format and the test DNS server gain NS, MX and TXT records, and simulation invents MX and TXT records (#194).
  • -takeover-providers <file> extends the built-in list of takeover-prone services behind provider: hints without waiting for a release: one <suffix> <name> pair per line, # comments allowed. An entry for a listed suffix replaces it, a name of - removes the suffix, and the longest matching suffix wins. -takeover-providers-only uses the file alone. Bad lines are a usage error naming the line. pkg/subenum gains Config.TakeoverProviders and Config.TakeoverProvidersOnly (#219).
  • Results whose addresses are not public are marked with the most notable scope: loopback, unspecified, link-local, private (RFC 1918, fc00::/7) or cgnat (100.64.0.0/10). A public name answering with one leaks internal addressing. The scope is an internal_address field in JSON and JSONL, INTERNAL=<scope> in text with -show-records, a tag in the TUI, and an internal_addresses count in the breakdown and the run-quality report; CSV is unchanged. Simulated results are never marked. pkg/subenum gains Result.Internal and Stats.Internal (#218).
  • -retry-inconclusive looks up once more, at the end of each scan, every name whose lookup timed out, was refused or failed, against -dns-server (never the -r pool). An answer replaces the first outcome in the counts, so a scan with a few dropped packets can still end complete; a name that resolves is reported like any other, but not expanded by -recursive. The pass is skipped after a reliability abort, an interrupt or once -max-duration runs out. -inconclusive <file> writes the names still without an answer, one per line, to rescan with -w. The report gains retried and recovered counts, and pkg/subenum gains Config.RetryInconclusive and Stats.Retried, Stats.Recovered and Stats.Inconclusive (#217).
  • Takeover hints recognize two more providers: a CNAME to *.animaapp.io is marked provider:anima and one to *.trydiscourse.com is marked provider:discourse (#134, thanks @jdotc1).
  • -log-format json writes the diagnostics on stderr (notices, warnings, errors, -v lines) as JSON Lines, one object per line with time, level (error, warn, notice, info, debug), kind (message, scan, verbose, progress, simulation, or an engine notice's category such as wildcard or cap), the target being scanned, the message, and the settings codes an engine message names (["max_queries"]). The simulation banner becomes one simulation notice, progress is an event with its numbers and only with an explicit -progress, and -silent keeps its meaning; results on stdout are unchanged. SUBENUM_LOG_FORMAT and the config file set it like any flag (#198).
  • -dry-run prints what a scan would do without scanning: per target the candidates left after dedupe, invalid entries and -exclude, the queries (one per name and record type, and the most with every -attempts retry, both capped by -max-queries; #223), how long they take at -rate, and whether the resolver passes the preflight, which is its only DNS query. -format json or jsonl prints the plan as JSON. No -o file is written, and the run exits 1 only if a preflight fails. -ct, -permute and -recursive add work it cannot count without querying, and it says so (#192).
  • pkg/subenum exports Stats.Verdict(interrupted) and the VerdictComplete, VerdictDegraded and VerdictUnreliable constants: a finished scan's verdict and reason by the same rules as the CLI's run-quality report. subenum mcp now takes its verdict from it rather than rebuilding the engine's counters by hand, and its reasons name max_queries and "the server's -max-duration" instead of raw setting tokens (#263).
  • pkg/subenum exports Permutations(seeds, words []string) []string, the candidate generator behind -permute: from names already found it returns the nearby variants (api-dev, dev-api, dev.api, api2), relative to the apex and ready to pass as a second scan's Words. nil words selects the CLI's built-in list. It sends no DNS queries (#197).
  • -tui pre-fills its form from the command line: subenum -tui example.com -w big.txt -r resolvers.txt opens with the domain, wordlist and resolver file filled in, over SUBENUM_*, config file and remembered values. Flags the form has no field for are named in a warning instead of being dropped silently; the form takes the first -w. The form gains Resolver Pool (-r) and Rate/Resolver (-rate-per-resolver) fields, remembered in last.json. After a scan, r restores the values just scanned rather than re-applying the launch flags over them (#195).
  • -rate-per-resolver N caps the DNS queries per second sent to each -r pool resolver, so a high global -rate no longer lands in uneven bursts on single members. Lookups go to resolvers with room first and wait only when every cap is spent; -rate still caps the total, and the stricter limit wins. The end-of-run resolver table and the report's resolvers entries gain a throttled count, and pkg/subenum gains Config.RatePerResolver (#196).
  • -w can be repeated to layer wordlists, for example a general list plus a target-specific one: -w general.txt -w target.txt reads both in order and scans an entry found in both only once (counted as a duplicate, as within one file). - stays valid as one of the entries. A w value from the config file or SUBENUM_W is replaced, not extended, by -w on the command line (#199).
  • -max-duration 20m gives a run a time budget. When it runs out no new names are tested, lookups already under way finish, and the run ends normally: results, the -stats report (stopped_by: "duration", untested names in skipped_by_cap) and a degraded verdict. Under -dL one budget covers every target. The GitHub Action gains a max-duration input, so a scheduled scan can stop before the job's timeout-minutes kills it without a report, and pkg/subenum gains Config.MaxDuration and Stats.Expired. With -max-queries, the report now says stopped_by: "queries" (#190).
  • The progress line and the TUI status bar show an estimated time left (| ETA 4m12s), from the rate over the last 30 seconds, once a scan has run for 5 seconds; it disappears at 100% (#191).
  • -diff-records makes -diff compare the records of names found again, so a CNAME re-pointed at an unclaimed service or a moved address is reported as changed (~ name in text; "change": "changed" with previous_records in JSON and JSONL), counted in the report's diff.changed, and exits 4. When either side has a CNAME only the alias targets are compared, so CDN address rotation is not a change. The previous file must be JSON, JSONL or CSV. The GitHub Action gains a changed output (#189).
  • The GitHub Action gains three outputs: changes, the path of the JSONL file listing the added, removed and changed names (empty when there were no previous results to diff against), so a later step can open an issue or post them; takeovers, the number of takeover candidates found; and internal, the number of names answering with an address that is not public. The job summary table has a column for each count and lists the takeover candidates under it. docs/monitoring.md shows a step that opens an issue from changes (#311).
  • The TUI's finished screen shows the run-quality verdict (complete, degraded or unreliable) with its reason above the counters, colored green, orange or red, so a scan that lost lookups or stopped at Max Duration no longer looks like a clean one. The counters line adds skipped candidates, retried and recovered names, takeover candidates and rate back-offs when there are any. The screen is the same height as before (#312).
  • pkg/subenum exports ProbeZone(ctx, domain, ZoneOptions), which tries a zone transfer (AXFR) from the domain's name servers with -axfr's limits (8 servers, 2 addresses each) and reports the names, the servers that allowed it and whether the zone uses plain NSEC (ZoneInfo), and ZoneTransfer(ZoneOptions), a Source for Config.Sources that scans the transferred names. Its queries are not paced by Rate, as -axfr's are not. The CLI and the library share the attempt in a new internal/axfr package (#314).
  • -dry-run takes the -r resolver pool into account: each pool resolver gets the target's preflight query too and has its own pool line (pool_preflight in JSON, with each resolver's address, preflight and ok), and the duration estimate uses -rate-per-resolver times the pool size when that is below -rate or -rate is unlimited, naming the flag that bounds it (rate_effective in JSON). A failing pool resolver does not fail the dry run, as a scan benches it and carries on (#376).

Changed

  • A -recursive scan over a wordlist of single labels no longer remembers every candidate it generated (about 75 MB per million), and one that reaches -max-queries or -max-duration counts the untested candidates without building each name, so it stops on time (#301).
  • A -recursive scan whose candidates include a multi-label name (a wordlist entry with a dot, or a nested name from -sources or -axfr) no longer remembers every candidate it generated either: it records only the names a multi-label entry begins, the only ones two parents can reach, so one nested name from CT no longer brings the per-candidate record back. In a benchmark of 100,000 entries plus one with a dot and ten found parents, the record drops from 1.1 million names to 11 and the scan allocates 26 MB instead of 135 MB (#359).
  • Smaller costs on large inputs: a -recursive scan with a multi-label wordlist entry and no -exclude that reaches -max-queries or -max-duration counts the untested candidates by walking only the entries that can repeat a name, instead of building every remaining name (100,000 entries under 200 found parents: about 2 ms and 0.2 MB instead of 2.2 s and 480 MB, so the scan stops on time); -axfr asks a target's name servers for the transfer at the same time instead of one after another, so servers that silently drop TCP/53 cost about one 10-second transfer timeout rather than one each (8 stalling servers in a test: 0.02 s instead of 16 s), with its messages still printed in server order; and subenum mcp holds at most 8 tool calls, running or waiting, per session and refuses the next with an invalid-request error, so a client pipelining calls with large arguments (up to 16 MB each) cannot grow the server's memory without bound (#365).
  • A run that -max-duration stopped with names untested, or that the reliability guard aborted, saves its resume state to the -state file like an interrupted one and prints continue with: subenum -resume <file>, so a time-boxed job continues where the last one stopped instead of testing the same first names again. The exit code is unchanged (0 for a spent budget). A resumed run that stops the same way saves its state again, and one that finishes no longer counts the names an earlier part skipped, so it can end complete (#294).
  • The command's source moved out of the repository root: the main_*.go files are now internal/cli/*.go, and the root main.go only calls cli.Main. go install github.com/TMHSDigital/subenum@latest, the Action, the Docker image and -X main.Version builds are unchanged. The lab scenarios are embedded through a small examples/labs package, and SECURITY.md is now .github/SECURITY.md.
  • A CNAME lookup sends one CNAME query instead of going through Go's LookupCNAME, which also sent A and AAAA queries and threw their answers away. -type A,AAAA,CNAME now costs three queries a name instead of five, so the same -rate tests two-thirds more names per second, and -dry-run, -max-queries help and the -rate-per-resolver reservation count it that way. The query goes through the same transports (UDP with TCP fallback, DoT, DoH), rate limiter and query counter as the other types (#249).
  • A name's record types are looked up at the same time instead of one after another, so with the default A,AAAA (or more types) a name costs its slowest answer rather than the sum of them, and a timed-out A no longer delays AAAA by the whole -timeout. -rate still caps the queries per second; results keep the order of -type (#250).
  • Passive sources (-sources, -ct, Config.Sources) retry a busy answer (429, 502, 503, 504) or a dropped connection twice, after 2 and 5 seconds or the server's Retry-After (at most 30 seconds), within each source's 60-second limit; crt.sh often answers 502 or 503 under load, which used to cost the scan all of its names. A target's sources are now asked at the same time instead of one after another (#271).
  • When the Wayback Machine's answer reaches the 200,000-URL or 64 MiB limit, a note says its results were truncated and some hosts may be missing, the report's sources entry (and the MCP report's) is marked truncated, explain_stats mentions it, and the library sends a NoticeSource event; the names it did return are still scanned. It used to stop silently (#272).
  • The members of an -r resolver pool share one budget of 512 idle UDP sockets, and a scan closes them when it ends, instead of each member keeping its own sockets for the whole run; a 10k-resolver list could hold 10k sockets open and hit file-descriptor limits (#252).
  • A target pasted with extra trailing dots (example.com..) or ending in a Unicode full stop (。, ., 。) normalises to the domain itself; the full stops separate labels as IDNA defines. It used to keep a trailing dot or turn the full stop into part of a label, and -exclude patterns with two trailing dots are now rejected instead of never matching (#281, thanks @bfs2021).
  • A Wayback Machine answer with one archived URL over 1 MiB skips that line instead of failing and losing every Wayback name (#259).
  • crt.sh's answer is read one entry at a time instead of decoded whole, so memory follows the names rather than the body. An answer that reaches the 64 MiB limit keeps the names read before the cut and is reported as truncated, like Wayback's, instead of failing and losing every CT name (#251).
  • The TUI runs on Bubble Tea v2: charm.land/bubbletea/v2 v2.1.1, bubbles/v2 v2.2.1 and lipgloss/v2 v2.0.6 replace the v1 modules, dropping v1-only helpers that had not had a release in years. Keys, screens and the form work as before; the go 1.26.0 minimum is unchanged (#188).
  • With text output (the default) and no -show-records or -diff, the CLI keeps only the names of the current target's results for -permute and the -resume state, plus the few results with a takeover hint, an internal address or a permutation, instead of every result with its records. A simulated 1M-name scan at 100% hits peaks at about 240 MB instead of 750 MB, against 233 MB at 1% hits. A resumed run re-emits the same text. Structured formats still keep the records a resumed run writes back (#183).
  • The scan dispatcher hands jobs to workers in small chunks (up to 8, fewer when there are few names per worker) over buffered channels, and each worker reports a chunk's jobs at once, instead of two unbuffered hand-offs per job. A 1M-name simulated scan at 50 workers (make bench) takes about 0.4 s instead of 1.6 s. A resumed scan may repeat up to one chunk per worker instead of one lookup per worker (#184).
  • Live UDP scans reuse sockets instead of opening and closing one per query, which capped throughput near 10k queries per second with nearly all CPU in socket teardown. A 100k-name scan against a loopback server now takes about 2 s instead of 12 s at -t 200. A late answer to an earlier query on a reused socket is ignored, as before, and cannot make a lookup count as refused (#180).
  • -format json streams its array instead of holding every result and then encoding the whole document, so its peak memory matches -format jsonl (a 300k-result run: about 200 MB instead of 420 MB). The output is byte for byte the same (#181).
  • The wordlist is normalized once per run instead of once per -dL target: wordlist.Prepare normalizes the lines and Prepared.For applies only each target's 253-character limit. On a 1M-line list, normalizing takes about 0.5 s instead of 0.8 s with almost no allocations, and each further target about 0.5 ms. Results and counts are unchanged (#182).
  • Tests no longer flake under load: lab end-to-end runs use a 5 s timeout with 3 attempts and fail with a clear message if the lab 2 wildcard probe times out, the resolver-pool spread test counts lookups per member instead of timing the scan (the timing view is now BenchmarkPoolSpreadsLoad), and CI runs tests in shuffled order (#167).
  • The names whose lookup timed out, was refused or failed are kept only when something reads them: for the -inconclusive file, and until the -retry-inconclusive pass has run. A scan without either only counts them, so under -no-abort or -auto-rate against a failing resolver memory no longer grows with every failed lookup (about 1 GB at 10M failures). When they are kept, a wordlist entry costs its index rather than its name. Counts, the report and pkg/subenum's Stats.Inconclusive are unchanged; the -resume state lists the names only with -inconclusive (#300).
  • Less memory on large wordlists and busy scans: -sources, -ct, -axfr and -permute check their names against the wordlist without building a set as large as it (about 200 MB per feature and target on a 5M-line list), a -simulate-zone lab scan no longer keeps a count per queried name (about 170 MB on a 1M-line list), and a CNAME lookup reads a UDP answer into a buffer sized for the 1232-byte EDNS payload it advertises instead of 64 KB. Results are unchanged (#321).
  • The TUI's charm.land/lipgloss/v2 moves from v2.0.6 to v2.1.1, which lowered its go line back to 1.26.0; Dependabot now skips only v2.1.0, the release that needed Go 1.26.7, instead of every later v2 release. The go 1.26.0 minimum is unchanged. The Docker image's org.opencontainers.image.licenses label is now GPL-3.0-only, the SPDX id CITATION.cff uses, instead of the deprecated GPL-3.0, and the example workflows in the monitoring guide pin the same upload-sarif and upload-artifact commits as the repo's own workflows, which a test now keeps in step (#366).

Fixed

  • A TUI scan with a Resolver Pool now writes the pool's per-resolver accounting (the resolvers block of the summary, with each resolver's benched, throttled and contradicted counts) to its JSON or JSONL output file, as -r does on the command line; the TUI built the summary with its own copy of the CLI's code, which had missed it. Both front ends now build the summary and its source rows with the same code (#367).
  • Lab mode (-simulate-zone) now answers custom scenarios as a recursive resolver would: an in-zone CNAME is followed through further CNAMEs (up to 8) and wildcards to the addresses at the end, where it used to stop after one step so a CNAME b, b CNAME www or a CNAME to a *.dev name read as missing; an IPv4-mapped AAAA value such as ::ffff:192.0.2.1, which the lab server never served, is rejected when the scenario loads; and a UDP answer larger than the query's EDNS buffer (512 bytes without one) is sent truncated so the client retries over TCP, where a large TXT value or many records used to fail the lookup. The shipped labs are unchanged (#371).
  • -exclude now keeps an out-of-scope target free of the queries sent before its scan, which used to go out before the scan refused it: -axfr no longer looks up its name servers, probes them or asks them for a transfer, -ct and -sources no longer ask about it, and -dry-run marks it out of scope (out_of_scope in JSON) instead of sending its preflight. -axfr also neither resolves nor contacts a name server that -exclude rules out, and the -dry-run preflights are paced by -rate. In pkg/subenum, a Domain that Exclude rules out gets no Sources lookup, and ZoneTransfer() skips excluded name servers (#372).
  • A -recursive scan stopped during its -retry-inconclusive pass, then resumed, no longer counts a name below depth 1 twice when the retry pass had recovered it. The part a resume repeats was tallied before the retry pass moved the name to its new outcome, so the saved counters kept it under that outcome while the resumed run counted it again, and the outcome counts no longer added up to processed; the tally now follows the name to its new outcome (#374).
  • A lookup is counted as refused only when the resolver answered with RCODE REFUSED or a DNS-over-HTTPS server shed load with HTTP 429 or 503. A socket error such as "connection refused", from a closed -dns-server port or a resolver that is down, used to be counted as refused too, so the verdict, the preflight error and the reliability guard pointed at rate limiting when the address was wrong; it is now counted as other, the same on every platform (#377).
  • A CNAME lookup no longer waits out its -timeout when a resolver ignores the advertised 1232-byte EDNS size and sends a larger UDP answer without setting TC. The read cut the answer, which then failed to parse and was skipped as a stray packet (on Windows the lookup failed as other); it is now asked again over TCP as if truncated. A matching UDP answer that fits but does not parse fails at once instead of timing out (#375).
  • The GitHub Action carries a time-boxed scan across runs. With max-duration and baseline-artifact, a run the budget stops saves its resume state in the baseline artifact, next to the previous results it still diffs against, and the next run continues that scan with -resume instead of starting from the first name again, so a large wordlist is worked through over several scheduled runs and the baseline is replaced once a run completes. The state is saved even when fail-on fails the step. A change to the action's inputs discards the state with a notice, as does subenum refusing to resume (for example a changed wordlist file), and a new scan starts. A new resumed output says whether the run continued a stopped scan. A later use of the action in the same job with the same baseline-artifact reads the copy the earlier use saved instead of the artifact list, which can lag an upload by seconds and made the later use start over or miss the baseline (#352).
  • The config file and SUBENUM_W/SUBENUM_O can set several -w wordlists or -o files, as the command line can. The config file takes a JSON array for w and o, one file per entry ("o": ["names.txt", "jsonl:results.jsonl"]), and for sources, as it already did for exclude and type; a string for w or o there or in the environment is a comma-separated list, as it is for the other list settings, where it used to become one path with commas in it. -print-config shows w and o as a JSON array, so one file whose name has a comma is told apart from two (#353).
  • A name whose lookup of one requested type timed out, was refused or failed while another type answered is no longer a plain found result with the records that came back. The whole lookup is tried again up to -attempts, and when the failed type never answers the result is marked partial: a partial field in JSON and JSONL, PARTIAL in text with -show-records, a partial answers line in the breakdown, and Stats.Partial in the library. Partial answers lower the run's verdict as failed lookups do (degraded at 1%), and -diff-records does not report a partial name as changed, since its missing records may only be a lookup that failed (#357).
  • Resuming a scan stopped in its -permute pass (Ctrl+C or -max-duration) no longer scans the finished main pass again. Under -recursive that repeated every lookup below the saved names, and a name the repeat found anew changed the permutation candidate list, so the resumed pass started at the wrong place in a different list, skipping some candidates and testing others twice. The resume state now records that the main pass finished, and the resumed run goes straight to the permutation pass with the saved names as its seeds (#358).
  • When -max-queries or -max-duration stops a scan whose wordlist holds multi-label entries (such as dev.api), the untested names are counted as they would have been: ones -exclude would remove count as excluded rather than skipped, so a run whose only untested names were out of scope stays complete, and under -recursive a name the rest of the scan would have reached through two parents is counted as skipped once, so tested, skipped and excluded add up to the candidate total (#361).
  • A CNAME target or name server that exists but has no A or AAAA record (NODATA, such as a name with only MX or TXT records) is no longer marked dangling or dangling-ns: Go reports it as not-found like a name that does not exist, so a target whose address lookup comes back empty now gets one more A query, and only an NXDOMAIN answer marks it. A probe that times out or fails marks nothing (#362).
  • A wordlist entry that already ends in the target, such as a full name pasted from another tool, is scanned as the prefix in front of it: www.example.com under example.com is www, not www.example.com.example.com, and one that repeats an entry already in the list counts as a duplicate. The scan notes how many entries it shortened (#351).
  • With -retry-inconclusive, names whose lookup failed before a run stopped (interrupt, -max-duration or the reliability guard), and that its retry pass never got to, are saved in the resume state and retried by the resumed run; one that then answers is reported and leaves the -inconclusive list (#315).
  • A run stopped by -max-duration, the reliability guard or an interrupt, then resumed, reports each name once: excluded is no longer counted by both runs, and processed, total and the outcome counts no longer include the lookups the resumed run repeats (a chunk reported after the stop, and under -recursive every lookup below the first level), so processed cannot exceed total. Names the resumed run looks up again also leave the saved inconclusive list (#341).
  • -axfr and ProbeZone say "no name servers found for " when the domain has no NS records or does not exist, instead of "name server lookup failed: ... no such host"; that wording was unreachable, since Go reports an empty NS answer as a not-found error (#347).
  • subenum -tui -ct puts ct in the form's Sources field instead of warning that the form has no field for -ct. explain_stats gives the permutation pass's own verdict, names tested and queries, and says when that pass set the target's verdict (#348).
  • The GitHub Action passes every line of a multi-line args: | block to subenum instead of only the first. fail-on: never no longer fails the step when a single-domain scan fails (exit 1 with an unreliable verdict); a usage error or a run with no report still does. A second use in the same job no longer reports the first use's verdict and counts when its own scan exits before writing a report. The found output and the summary table count -permute hits too (#363).
  • The GitHub Action's job summary cuts its takeover and change lists to 50 entries in jq itself instead of piping into head -50, which under the step's pipefail could fail the step on a very long list after the scan had succeeded (#345).
  • The TUI's finished screen wraps its counters between counters to the terminal width and gives the extra lines to the layout, so with every counter set it still fits an 80×24 terminal instead of the terminal wrapping the line and pushing the screen one row down. ctrl+c during the permutation pass after a -no-abort reliability warning shows an abort, not a failure (#343).
  • Standard input can be named by only one input flag: -w -, -dL -, -r -, -exclude-file - and -takeover-providers - are all counted, and a second one is a usage error naming them, before anything is read. Only -w with -dL used to be refused; with the others the first reader drained standard input and a later one failed on an empty list, with an error about that list instead of the shared input. One -o file named twice in different spellings (a relative and an absolute path, x and sub/../x, another case on Windows and macOS), or an -inconclusive, -stats or -sarif file that is also an -o file or another of them, is a usage error too, where only identical spellings were refused and the last file written silently replaced the others. A subenum mcp scan call that cannot start (an unsupported type, an invalid exclude pattern, no usable words) is reported not_run as before but no longer spends one of the session's -max-calls (#378).
  • A -diff baseline that is also an -o file is recognised however the two paths are spelled (x and ./x, a relative and an absolute path, another case on Windows), so a stopped run leaves it unchanged for the resumed run, as it already did for identical spellings (#342).
  • A CNAME lookup of a name with an empty label or one over 63 bytes is reported as not found without a query, as a name over 255 bytes already was, instead of dialling, counting a query and reporting a failure that the reliability guard counts (#344).
  • Under -no-abort, the reliability guard's warning is printed as Warning: ... (level warn with -log-format json) instead of Error: warning: ..., and a target that finished no longer carries it as its error in the -stats report next to "status": "ok". The TUI shows it as a warning too (#308).
  • The resolver preflight and wildcard probes made before a target's first name no longer use up -max-duration. Under a low -rate they could take all of a short budget, so the run tested no names and each -resume probed again and made no progress. A budget that runs out before any name was tested now says so (#340).
  • Release archives and the container image carry the licence notices of the Go standard library and every module linked into the binary, in a THIRD_PARTY_LICENSES file generated at release time by tools/licenses from the module cache. The image also carries the project's LICENSE; both are in /usr/share/doc/subenum/ (#326).
  • With -format csv, each row reaches stdout as its result is found, as with the other formats, instead of when 4 KB had built up or the run ended; a pipe sees live rows and a force quit loses none (#319).
  • With both -max-queries and -max-duration, a deadline that fires after the query cap no longer loses the chunk of names admitted but not yet handed out: they count as skipped, so tested plus skipped adds up to every candidate and progress reaches 100%, and the run is marked as stopped by the time budget. Children of a name found after a budget stop whose whole branch is excluded count as excluded, not skipped, so they cannot make a complete scan look incomplete (#317).
  • A -r resolver list line with a malformed or out-of-range port (9.9.9.9:abc, :99999, :0) is rejected when the file is read, naming the line, instead of failing every lookup at dial time; an explicit -attempts 0 (flag, SUBENUM_ATTEMPTS or config) is rejected like any value below 1 instead of running as 1, and -attempts 0 -retries 3 reports that both cannot be used (#318, thanks @bfs2021).
  • Takeover hints no longer look up a CNAME target or name server that -exclude puts out of scope, so excluded names are never queried on this path either. Such a target can still be marked provider:<name>, which needs no query, but never dangling or dangling-ns (#316).
  • The TUI configuration form fits a terminal shorter than its 31 lines, such as the standard 80×24: the logo is dropped, then the field rows scroll to keep the focused field in view, with ↑ N more fields and ↓ N more fields lines for the rest. The validation error and the key hints are always shown, so a rejected ctrl+r no longer fails without a visible reason (#287).
  • The run-quality report counts the -permute pass: a target's and the run's queries_sent, and so achieved_qps, left its queries out, often most of the traffic sent. A permutation pass that went worse than the main pass (failed lookups, a spent -max-queries or -max-duration budget, a reliability abort, or a pass that could not start) now sets the target's and the run's verdict, with a reason starting permutation pass:, so a job gated on the verdict no longer passes on it. The report's permutation entry gains the pass's own processed and queries_sent (#298).
  • A run with -diff and -o on the same file that is interrupted (or stopped by -max-duration or the reliability guard) and saves a resume state now leaves that file as it was, instead of replacing it with its partial results. The resumed run compared against the partial file, so it reported names that were already known as added, exited 4 with nothing changed and could not report removals. The partial results are in the resume state, and the resumed run writes the full file; -resume now also refuses to continue if that file changed in between (#297).
  • A resumed -recursive scan checks each name found before the interrupt for wildcard DNS before expanding it, as a live scan does. A wildcard subzone, or one whose check was inconclusive, is skipped with the same notice instead of reporting every wordlist entry under it as a result, so an interrupted-then-resumed scan reports the same names as an uninterrupted one (#296).
  • A resumed -recursive scan expands a name found through a multi-label wordlist or passive-source entry (such as dev.api) as deep as an uninterrupted run does: its recursion level was taken from its label count, so its children were skipped or cut one level short. A resumed -dL run's report keeps the -permute pass, passive-source counts and -axfr details (axfr_allowed, axfr_names, zone_denial) of targets that finished before the interrupt, and those targets no longer fetch passive sources or try the zone transfer again, nor lose those details from the state when the run is interrupted once more (#315).
  • The TUI no longer requires a wordlist file: an empty Wordlist field scans with the bundled top-5000 list, as a missing -w does on the command line, and the scan view says which list ran. The field no longer pre-fills examples/sample_wordlist.txt, which only exists inside a checkout, so subenum -tui from an installed binary works on the first run (#288).
  • On Windows, -o NUL, -stats NUL and -sarif NUL no longer fail with "Incorrect function" and exit 1: device and pipe targets are opened without truncation (#208, thanks @bfs2021).
  • An unreadable -seeds file no longer leaves a hidden .<name>.tmp file beside the -o path (kept locked on Windows until exit): seeds are read before the output file is opened (#212).
  • On Windows, -simulate-zone labs and the test DNS server no longer fail to start about a quarter of the time with "no free UDP+TCP port pair": after a failed TCP bind the server retries on random ports instead of the next OS-assigned one, which stayed inside the same reserved port block (#204).
  • A scan.Config built directly with Attempts: 0 now makes one attempt per name instead of passing preflight and then failing every lookup without a query. Scan workers, scan.Verify and the wildcard near-miss probes share one resolve path (test hook, simulation, or the trusted resolver), and the internal/dns lookup helpers take a dns.Lookup value (resolver, timeout, attempts, types, logger) instead of long positional parameter lists, so a swapped attempts/types argument no longer compiles (#163).
  • -rate-per-resolver (and pkg/subenum's RatePerResolver) now holds: each pool lookup books one slot per DNS query it sends, as -rate does, instead of one per record type. The default A and AAAA lookup from the library booked one slot for two queries, and a CNAME lookup one for three, so members got up to twice the cap (#236).
  • -diff's check of previous names this run did not find now stays within the run's budgets: it looks up at most what is left of -max-queries (per target, less the names tested) and stops at the -max-duration deadline, counting the names it does not reach as unverified. A large baseline could otherwise send far more queries, and run far longer, than the budget allowed (#276).
  • Shell completion for -type offers NS, MX and TXT, which -type has accepted since #194. The record types and output formats now come from one list each, which completion, the MCP schema and flag checking share (#264).
  • When -max-queries or -max-duration stops a scan, candidates -exclude would have kept out are counted as excluded, not skipped, so a budget that ran out only past excluded names leaves the verdict complete (and -diff reporting removals) instead of degraded (#238).
  • -resume no longer skips names when a passive source failed in the interrupted run and works on resume. The failed source was fetched again, its names were inserted before later sources' names, and the saved position then pointed past candidates that were never tested while the verdict could still read complete. The resume state now records a failed source with no names, and the resumed run replays that (#237).
  • -resume asks a passive source (-ct, -sources) again, and repeats the -axfr zone transfer, when the interrupt arrived during that lookup, before the target's scan had started. The cut-short lookup was saved as a final, empty result, so the resumed run scanned without those names and reported 0 names. A lookup that failed in a scan that had started is still replayed as saved (#299).
  • An A or AAAA lookup that timed out no longer sends the same name again in the background. The repeats went out about five seconds apart on a reused socket, outside -rate and -rate-per-resolver, uncounted in the query total, and could continue after the scan had ended; a lookup whose time is up now gets no connection (#295).
  • The docs no longer present -type NS as available in the latest release. The site tags NS, MX and TXT as unreleased (new values of a released flag can now be tagged), the monitoring guide's code-scanning example no longer adds NS to a workflow pinned to a release that rejects it, and the README says its flag table describes main (#306).
  • subenum -h lists subenum diff among its usage lines, as it already did its exit code, and the docs catch up with the code: the monitoring guide's issue step is skipped rather than failing when the pinned release has no changes output, the README and Docker page mark -sources, -axfr, subenum diff and subenum mcp as unreleased, the library's MaxDuration says the preflight and wildcard probes are not counted, and the architecture notes and developer guide match the current packages, CSV columns and resume state (#379).

Security

  • go.mod names a patched toolchain (toolchain go1.26.11) beside the unchanged go 1.26.0 floor, so go install and the action's version: source build with the security-fixed net/http, crypto/tls and crypto/x509 instead of the original 1.26.0 standard library; with the default GOTOOLCHAIN=auto an older local Go fetches it. Release binaries were already built with a current Go (#243).
  • The GitHub Action checks the downloaded release archive's build provenance with gh attestation verify as well as its checksums.txt, since an archive and its checksums file could be replaced together. The Pages workflow grants pages: write and id-token: write only to the job that deploys, not to the whole workflow (#210).

0.10.0 - 2026-11-08

Changed

  • Engine messages no longer hard-code command-line flags. They name settings with neutral tokens (scan.Setting), and each front end renders them in its own terms: the CLI as flags (unchanged output), the TUI as its form labels ("Use Force to scan anyway", "invalid Exclude pattern"), and pkg/subenum as Config fields, replacing its regexp that guessed flags out of message text. A test fails if an engine message names a flag or uses an unknown token (#100).
  • The scan engine always ends with exactly one done event, marked Stopped when the scan ended before testing any candidate (out of scope, recursion ceiling, failed preflight or wildcard check, a wildcard zone without -force, or Ctrl+C during those checks), so the CLI, the TUI and pkg/subenum handle one shape of ending instead of a channel that sometimes closes without one. pkg/subenum's Event gains Stopped, and a stopped scan's Stats now include the queries its checks sent (#99).
  • -retries' deprecation warning goes through the normal output, so -silent hides it (#166).
  • The container image is built with Go 1.27.3 (from 1.27.1, picking up crypto/tls and crypto/x509 fixes the -ct client uses) on a refreshed distroless base (#185). golang.org/x/net is v0.62.0 and indirect modules are on their latest patches; the go 1.26.0 minimum is unchanged (#186).
  • CI pins govulncheck and the devcontainer images, which Dependabot now updates, and lints with golangci-lint v2.15.0 (#187).

Fixed

  • With -r, a retry goes back to the healthy resolver it just used rather than to one that is benched as dead, so a pool with one live resolver no longer turns transient timeouts into failures (#148).
  • pkg/subenum: Scan with NoAbort returns no error for a scan that ran to completion; the reliability warning is still sent by Run as a KindError event (#149).
  • -simulate no longer contacts crt.sh when -ct is set: CT seeding is skipped with a note, keeping the promise that nothing leaves the machine (#150).
  • -diff interrupted while re-checking previous names counts the names it never checked as unverified instead of still resolving (#151).
  • -exclude accepts internationalized patterns such as mail.bücher.de and matches them in punycode, as targets are scanned (#152).
  • pkg/subenum: a simulated scan with Seed 0 picks a random seed, as documented, instead of always using 0 (#154).
  • A resumed scan counts names it finds again after the resume point only once in found and processed (#156).
  • The TUI keeps reading scan events when the engine sends a kind it does not know, instead of stalling (#165).
  • -resume= with no path is a usage error, and -resume after -- is treated as an argument (#166).
  • The TUI exits 1 when the results file could not be written, as the CLI does, instead of 0 with only an on-screen message (#155).
  • -print-config shows the attempts value a scan actually uses: 1 by default instead of the raw 0 sentinel, and the -retries value (marked from retries) when that is what wins (#153).
  • tools/wordlist-gen parses flags that appear after a positional the same way subenum does, instead of silently ignoring both: a leftover argument is now a usage error (2) that points at -domain, and an invalid -combine value or an empty -o exits 2 instead of 1 (#120).

0.9.0 - 2026-10-06

Added

  • A GitHub Action: uses: TMHSDigital/subenum@<release> installs a release verified against checksums.txt, diffs against the previous complete run (kept as an artifact, or a previous file), can write SARIF, writes a job summary table, exposes verdict/found/added/removed outputs and fails the job on a configurable verdict. CI runs it from the checkout against a lab zone on Linux, macOS and Windows (#126).
  • Certificate Transparency seeding: -ct adds the names crt.sh knows under each target (one HTTPS request, no API key) as candidates after the wordlist and as -permute seeds; every name is still resolved before it is reported. A crt.sh outage is a warning, the report gains ct_names and ct_error, a resumed scan reuses the fetched names, and lab mode refuses -ct (#130).
  • -sarif <file> writes takeover candidates as a SARIF 2.1.0 log (validated against the OASIS schema): one rule per kind of finding, with a security severity, the CNAME target in the message and a stable fingerprint per name, ready for github/codeql-action/upload-sarif. The Monitoring page shows the upload step (#129).
  • -silent for pipelines: bare result names on stdout and nothing on stderr but errors and a one-line warning when results are simulated. A new Pipelines page chains subenum with dnsx, httpx and nuclei and gates each step on the run-quality verdict (#128).
  • A Go library API, pkg/subenum, so other programs can embed the engine: Config (every zero value takes the CLI's default), Scan for a finished run, Run to stream Event values, and Result, Record and Stats. The DNS internals stay unexported, engine messages name Config fields instead of CLI flags, and Run always ends with one terminal event carrying the final Stats. Runnable examples ship for pkg.go.dev (#77).
  • Lab mode: -simulate-zone <file> answers every query from a scenario file (names, A/AAAA/CNAME records, wildcards, TIMEOUT/SERVFAIL/REFUSED/NXDOMAIN failures, $DELAY and $REFUSE-ABOVE rate limiting) through a DNS server on 127.0.0.1. The real resolver, wildcard detection, -rate, takeover hints and the run-quality report run unchanged, deterministically and without traffic leaving the machine. A new Labs page has five guided exercises with answers, and the scenarios ship in examples/labs/. The run-quality report gains a zone field (#76).
  • Contributor experience: a devcontainer (Go 1.26 plus the CI golangci-lint) with an "Open in Codespaces" badge, YAML issue forms with blank issues disabled and contact links to private vulnerability reporting and Discussions, FUNDING.yml and CITATION.cff (#75).
  • Encrypted DNS transports: -dns-server tls://host[:853] (DNS over TLS, with pooled connections) and -dns-server https://host/path (DNS over HTTPS, RFC 8484, keep-alive and HTTP/2). Records, outcome classification (including REFUSED and timeouts) and -rate accounting are identical to UDP, and the run-quality report gains a transport field (#86).
  • Resumable scans: an interrupted run (Ctrl+C, SIGTERM) saves its state to subenum-resume.json (or -state <file>), and subenum -resume <file> replays the original command line, checks the input files are unchanged, restores the earlier results and continues from the first wordlist entry not yet fully looked up. The scan engine tracks that low-water mark exactly even though workers finish out of order, so at most a worker pool's worth of lookups is repeated (#73).
  • Permutation mode: -permute follows each target's scan with a second pass over permutations of the names found (word joins such as api-dev, dev-api and dev.api, and number increments), and -seeds <file> adds seeds from an earlier run in any format. The pass is fully guarded (preflight, wildcard filtering, -rate, -max-queries, -exclude), skips wordlist entries, marks its hits "permutation": true, and reports seeds, candidates and hits in the run-quality report (#72).
  • Defaults from a config file and the environment: config.json in the user config directory (or $SUBENUM_CONFIG) and SUBENUM_<FLAG> variables, with precedence flag > env > config > built-in. -print-config shows each effective setting and its source. subenum completion bash|zsh|fish|powershell and subenum man generate shell completions and a man page from the flags, and release archives ship them (#88).
  • A useful default wordlist: -w is now optional. Without it, subenum uses an embedded list of the 5,000 most common subdomain labels (SecLists subdomains-top1million-5000.txt, MIT; the license notice is kept in data/subdomains-5k.txt) and says so on stderr. The README gains a "Choosing a wordlist" section on larger lists and their query cost (#63).
  • Monitoring mode: -diff previous compares with an earlier results file (any format) and prints only names added (+ / "change": "added") or removed (- / "removed"), with a change CSV column; exit code 4 signals changes. The -o file keeps the full current results for the next run. Removals are reported only when the run's verdict is complete, and never for -excluded names. The run-quality report gains a diff object, and docs/monitoring.md has a scheduled GitHub Actions workflow that opens an issue for new names (#84).
  • Subdomain takeover hints: results with a CNAME are marked dangling (target is NXDOMAIN), provider:<name> (target at a takeover-prone service such as S3, GitHub Pages, Heroku or Azure) or dangling:<name>. The marker appears in JSON/JSONL (takeover_candidate), as a CSV column when CNAME records are requested, in text with -show-records, in the TUI, and as a count in the breakdown and report. DNS-only; no HTTP probing (#71).
  • Resolver pool: -r resolvers.txt spreads lookups over many resolvers, round-robin over the healthy ones (a resolver failing over half its recent lookups is benched for 30s), with retries sent to a different resolver. Every pool hit is re-validated against -dns-server, whose answer is reported, so a lying pool resolver cannot inject results. Per-resolver accounting appears in a stderr table and in the run-quality report, with pool_hits / confirmed / unconfirmed per target. In a hermetic test, 20 rate-limited resolvers ran the same scan 14x faster than one (#69).
  • Scope control: -exclude 'vpn.example.com,*.corp.example.com' and -exclude-file keep out-of-scope names out of a scan. Excluded names are never queried, wildcard-probed or expanded recursively (a found parent whose whole subtree is excluded costs no queries), they are counted as excluded in the breakdown and the run-quality report, and a target that is itself out of scope is refused. The TUI has a matching Exclude field (#87).
  • Run-quality report: -stats <file> writes a schema-versioned JSON summary of a run (outcomes per target, DNS queries actually sent, achieved query rate, wildcard fingerprint, candidates skipped by -max-queries) with a complete / degraded / unreliable verdict and its reason. -format jsonl ends with the same object as its last line. The README documents the schema and how to gate CI on the verdict (#70).

Changed

  • The site publishes a sitemap.xml (jekyll-sitemap) and a JSON-LD SoftwareApplication block on the home page; it redeploys only when docs/ changes or a release is published, and the build fails on a broken internal link or #anchor (.github/scripts/check-links.py) (#125).
  • The website is redesigned and rebranded: a new mark and wordmark (an apex branching into resolved and NXDOMAIN names), a palette built from the outcome colours, a landing page that replays a scan and explains the three verdicts, and a docs layout with a sidebar, on-page contents, previous/next links, client-side search (/ or Ctrl+K), light and dark themes and a 404 page. New pages: Getting started, CLI reference (the flag table moves off the home page) and Go library. Navigation is driven by docs/_data/nav.yml; the Developer Guide explains how to preview the site locally.
  • Releases are built by GoReleaser from .goreleaser.yaml: static, trimmed binaries for Linux, macOS, Windows and FreeBSD (adding windows/arm64, linux/386 and FreeBSD) packed in .tar.gz/.zip archives with LICENSE, README and CHANGELOG, plus checksums.txt, an SPDX SBOM per archive, a keyless cosign signature over the checksums, build-provenance attestations, and release notes taken from the tag's CHANGELOG section. The release stays a draft until provenance is attested. CI runs the same pipeline in snapshot mode on every push to main. Asset names change from subenum-linux-amd64 to subenum_<version>_linux_amd64.tar.gz (#61).
  • The GitHub Pages site is current: the hero shows real output (resolved names only), the flag table is generated from the CLI's own flags (a test fails when it drifts; make docs-flags regenerates it), the features cover the run-quality report, resolver pool, takeover hints, monitoring and scope control, there is a GHCR quick start and a Monitoring page, shared links get og:/Twitter preview cards with a new social image, the internal DOCUMENTATION_STRUCTURE.md template is no longer published, and the unused Cayman theme is gone. The README's "Why subenum" is updated to match (#64).
  • Tests run against a shared, scriptable DNS server, internal/dnstest, built on golang.org/x/net/dns/dnsmessage. It replaces the two hand-encoded responders, answers per name and type with records, any RCODE, NODATA with SOA, delays, drops or truncation, and counts queries per name and type. New tests cover NODATA, a resolver at an IPv6 address, exact wire queries per record type, and TCP fallback after truncation (#67).
  • The minimum Go version is now 1.26 (go 1.26.0 in go.mod), required by current golang.org/x/sys and golang.org/x/text; both were bumped and govulncheck -show verbose now reports no vulnerable modules. Dependabot also updates indirect modules, and CI lints with errorlint, nilerr, contextcheck, noctx, copyloopvar, usestdlibvars and revive, with errcheck exemptions narrowed to stdout prints (#83).

Fixed

  • The TUI writes the same files as the CLI: a CSV gets the takeover_candidate column when CNAME records are requested, and a JSONL file ends with the run-quality summary line. Its form starts from SUBENUM_* and config-file values, over the values remembered from the last session. The report is built by a new internal/report package that both front ends share (#118).
  • Shell completions: -simulate-zone, -seeds, -resume and -state complete file names in every shell (fish previously disabled file completion for them), bash and fish complete the shell name after completion, and the PowerShell completer reads the previous word, so file flags fall back to path completion and -format/-type offer their values. A test fails when a flag that takes a file is missing from the list, and CI checks that every generated script parses in its shell (#117).
  • New -o and -stats files get 0666 less the umask, as os.Create would, instead of a fixed 0644, so results stay private under umask 077; an -o path that is a symlink keeps the link and replaces the file it points at, instead of turning into a regular file (#119).
  • Takeover hints resolve each CNAME target once per scan instead of once per result, so hundreds of names aliased to the same CDN or SaaS host no longer spend a query and a -rate slot each; concurrent results for one target share the lookup, and an inconclusive one is retried by the next name (#121).
  • DoT connection reuse is accurate: a pooled connection the server already closed is detected before reuse instead of failing a query that had already been counted and paced (a server closing after each answer cost 19 dials for 10 lookups; now 10). Up to 128 connections are kept, matching the default -t, and a scan closes its idle DoT and DoH connections when it ends, so a program that scans repeatedly through pkg/subenum does not hold them open (#113).
  • DNS over HTTPS never follows a redirect, so a misconfigured endpoint or proxy cannot move the query elsewhere or downgrade it to cleartext http://; HTTP 429 and 503 count as refused (the rate-limit signal) instead of other (#114).
  • -dns-server tls://[2001:db8::1] (a bracketed IPv6 host without a port) is accepted; the validator and the dialer now share one parser. Lab zone CNAME targets are validated, and CNAME @ points at the scanned domain instead of the bogus @.<domain> (#115).
  • The test DNS server's Close waits for every handler, closes open stream connections, and no longer returns while delayed UDP answers or connections parked on a dropped query are still running. New tests cover an untrusted DoT certificate, DoH redirects and throttling, stale pooled connections, a lying pool member and the server's own lifecycle (#122).
  • pkg/subenum.Run no longer leaks the scan's worker pool when a caller breaks out of the event loop and cancels without draining (about 100 goroutines per run before): after cancellation the forwarder never waits for a reader, so every goroutine exits within about a second. DefaultWords parses the bundled list once, and a positive Timeout below 1ms is rounded up instead of being rejected as 0 (#112).
  • A resolver pool member that answers NXDOMAIN for names that exist can no longer silently hide them. Each member is checked with canaries (names the trusted resolver confirmed, starting with the apex), and an NXDOMAIN from a member that has not passed a canary is confirmed by a second member. A member contradicted three times (a hidden name, a failed canary, or a hit the trusted resolver denies) is flagged and dropped for the rest of the run; the per-resolver report gains canaries, contradicted and flagged. In a hermetic test, one always-NXDOMAIN member previously hid half the real names; now none are lost. In a zone with no known name to use as a canary, NXDOMAIN answers cost a second pool query until one is found (#105).
  • -diff no longer reports names this run never tested as removed. Every previous name the run did not find is looked up again with the trusted resolver, and only NXDOMAIN counts as removed; names that still resolve (a smaller wordlist, no -recursive or -permute, now wildcard-filtered) or whose lookup fails are counted as still_resolving and unverified in the report's diff object instead (#110).
  • Resuming a -permute scan no longer skips the permutation pass: the main pass's resume position was applied to the permutation candidates, so a resumed pass queried nothing. The pass now saves and resumes from its own position, and is seeded only with the main pass's names so the resumed candidate list is identical (#107).
  • -max-queries covers a scan across interrupts: a resumed pass gets only the budget left, and none once it is spent. Previously each resume started a fresh budget (#108).
  • Resume state is more robust: a resumed scan interrupted again during preflight keeps the earlier resume point and statistics instead of restarting from entry 0, targets a run never reached keep their state, and a resume is refused when the normalized wordlist (including the bundled one), an effective setting from SUBENUM_* or the config file, or a -simulate-zone, -seeds or -diff file changed. A resumed run that finishes deletes its state file (#109).
  • -permute with -recursive runs the permutation pass one level deep, instead of expanding found permutations with the candidate list (names such as <candidate>.api-dev.<target>) and being refused past the recursion ceiling; candidates longer than 253 characters are skipped instead of failing as other (#111).
  • Lab mode can no longer send queries off the machine: -r, -dns-server or -simulate from a SUBENUM_* variable or the config file is refused with -simulate-zone just like the flag, naming where the value came from. Previously SUBENUM_R sent every lab query to the listed resolvers under a "no traffic leaves this machine" banner (#106).
  • Config and environment defaults: -h, -version, -print-config and -tui work despite a bad value; every bad value is reported at once; config values must be strings, numbers or booleans (exclude and type also take a list) instead of being stringified into values such as <nil>; resume is refused in the config file; -attempts and -retries from different sources resolve by precedence instead of conflicting; and SUBENUM_PROGRESS counts as an explicit -progress (#116).
  • An unreadable -r file exits 1 like every other unreadable input file (a malformed one is still a usage error, 2), and argument errors are reported before the simulation banner is printed or any resolver list is read (#120).
  • A wildcard check whose probes time out or SERVFAIL is no longer read as "no wildcard". Probes are retried up to -attempts; if still inconclusive, the scan aborts (scans without filtering under -force, with a warning), and in recursive mode the branch is skipped and not cached. Previously a wildcard zone with flaky probes reported every wordlist entry as a hit (#47).
  • The dispatcher no longer does quadratic queue copying, so large wordlists scale linearly (simulated 2M entries: 19.4s down to 4.6s). Candidate names are generated lazily from the wordlist instead of being built up front, and a found parent is expanded in the dispatcher instead of pushing every child through a channel. make bench runs a 1M-entry benchmark (#49).
  • -o results files are written to a temporary file beside the target and renamed into place only once a scan finishes (including an interrupted one with partial results), in both the CLI and the TUI. A run that fails before scanning (preflight failure, wildcard abort, recursion refusal, Ctrl+C during preflight) no longer truncates an existing results file (#52).
  • Progress no longer corrupts results when stdout and stderr are merged (2>&1, tee, CI logs). Progress is off by default when stderr is not a terminal, an explicit -progress there prints whole lines with no carriage returns, and an active progress line is always ended before a result is printed (#78).
  • Simulated results identify themselves wherever they are saved: JSON and JSONL objects carry "simulated": true, CSV gets a simulated column, -o text files start with a # SIMULATED ... -seed N comment line, and the CLI ends a simulated run with a stderr reminder. Piped bare names stay unmarked (#53).
  • The preflight lookup of the target domain is retried up to -attempts, so one dropped packet no longer aborts the scan (#57).
  • -max-queries now stops memory growth in recursive mode: once the cap is reached, remaining and future candidates are counted as skipped without being built or recorded (#51).
  • No default command sends DNS traffic at a domain you did not name. docker compose up, make simulate* and examples/demo.sh run in simulation mode; make run, run-verbose, run-custom and docker-run require DOMAIN= and stop with a message without it; demo.sh -l <domain> opts into a live run. demo.sh now uses set -euo pipefail, docker-run uses $(CURDIR) so it works under Windows make, and .PHONY lists every target (#81).
  • Target domains are normalized the same way in the positional argument, -dL and the TUI: trimmed, lowercased, trailing dot dropped, so Example.COM. no longer yields api.Example.COM.. A pasted URL, path, :port or leading *. is stripped with a one-line note, IDNs such as bücher.de are converted to punycode, underscores are accepted as in wordlist entries, and an email address gets a targeted hint instead of invalid domain format (#54).
  • TUI: a - wordlist is rejected on the form instead of freezing the UI on stdin, the wordlist loads in the background with a "Loading wordlist..." line so large lists no longer block input, a scan that ends with an error (preflight failure, wildcard abort, reliability abort) shows "Failed: " instead of "Aborted", and the TUI exits 1 when the last scan failed (#55).
  • -rate now charges every DNS query that reaches the wire. A CNAME lookup costs three slots (Go sends A, AAAA and CNAME queries), and queries Go's resolver retries internally on SERVFAIL, REFUSED or socket errors are charged as they are dialed, so a struggling resolver no longer gets more traffic than -rate allows. Responses with RCODE REFUSED are now counted as refused instead of other, and lookup errors name the configured -dns-server instead of the system nameserver (#50).
  • Wildcards that answer from a rotating CDN or load-balancer pool are now filtered under -force. The root fingerprint starts from five probes instead of two and, when their answers differ, keeps probing until the pool stops growing; results that share only some records with the fingerprint are re-validated with fresh probes under the same parent. In a hermetic 8-address pool test, filtering went from 12 of 50 wildcard answers to all 50, with no real names lost (#48).
  • A wordlist line longer than 64 KB no longer aborts the run with bufio.Scanner: token too long. Overlong lines are truncated, skipped and counted with the other invalid entries (#56).
  • tools/wordlist-gen no longer overwrites an existing output file without the new -f flag, exits 1 instead of writing an empty file when nothing is generated, rejects -combine prefixes that are not valid DNS labels, and drops generated entries subenum would skip (for example combinations over 63 characters). make wordlist and examples/demo.sh pass -f for the files they own (#59).
  • CLI polish: -h shows the -attempts default, a missing -w or domain is named before the usage text, -version prints before the simulation banner, -tui is only honoured as a flag (not as -o -tui or after --) and warns that other flags are ignored, and the worker pool is capped at the number of jobs the scan can produce, with a warning above -t 10000 (#58).
  • Exit codes are consistent and documented in the README and -h: 2 for every usage error (previously 1 for semantic ones such as a missing domain or bad -format), 3 when some -dL targets failed while others completed, and 143 instead of 130 for SIGTERM, without the Ctrl+C hint. With -dL, a reliability abort now skips the remaining targets instead of sending them to the same overloaded resolver (unless -no-abort), and a per-target status list is printed at the end (#82).
  • The CLI and the TUI now share one scan.Options validator and scan.Config builder, so they accept the same values and build identical scans. The TUI now shows the simulation seed (so runs can be reproduced) and the number of skipped wordlist entries, skips the resolver check in simulation mode like the CLI, and saves its settings atomically. The default resolver is a single constant. The CLI no longer rejects an unused -hit-rate outside simulation mode (#80).
  • Release binaries are built with CGO_ENABLED=0 -trimpath -s -w, so subenum-linux-amd64 is statically linked again and CI fails if a Linux asset is not (#46).

0.8.0 - 2026-10-04

Fixed

  • Flags written after the domain (subenum -w wl.txt example.com -t 50) are now parsed instead of silently ignored. Extra positional arguments are rejected, -h exits 0, and unknown flags exit 2 (#29).
  • -type CNAME no longer counts names that exist without a CNAME as resolver failures (other). A successful lookup with no records of the requested types is now a definitive negative (counted under nxdomain, not retried), so it cannot trip the reliability guard. Wildcard probes now use the requested record types, so CNAME wildcards are fingerprinted and filtered under -force (#28).
  • Lookups cut short by Ctrl+C, SIGTERM, or the reliability guard's own cancellation are no longer counted as failures, which previously inflated other and could raise a false "likely resolver rate-limiting" abort with exit 1. An interrupted CLI scan now exits 130 after flushing partial results (#42).
  • Output-file write failures (disk full, closed handle) now make the CLI exit 1 instead of 0, and the TUI shows an "output file incomplete" error instead of discarding it. Writer.Finish returns the first file write error, including CSV writer errors. The output file is created only after the wordlist loads, so a bad -w path no longer truncates an existing -o target (#30).
  • -rate now bounds DNS queries on the wire instead of candidate names. Every record type (CNAME counts as two), retry, wildcard probe and the preflight takes a slot, so -rate 50 with the default A,AAAA no longer sends up to 100 qps. A reservation-based dns.RateLimiter replaces the ticker, which dropped ticks and under-delivered under load (ROADMAP N1). Time spent waiting for a slot does not count against -timeout (#32).
  • Each record type now gets its own -timeout, so a slow A answer no longer starves the AAAA/CNAME lookups that shared its budget (ROADMAP N3).
  • ResolveTypes reports the most severe per-type error instead of the last one. An A SERVFAIL or timeout followed by an AAAA NXDOMAIN is now a failure that -attempts retries and the reliability guard sees, instead of a silent NXDOMAIN miss (#31).
  • Lookups use absolute names (trailing dot), so resolv.conf search suffixes (for example Kubernetes ndots:5) no longer multiply the queries sent for every missing name.
  • TUI scan view: the viewport height is clamped to at least one row, so terminals shorter than the layout no longer produce a negative height (#37, ROADMAP N6). The layout reserve also accounts for the hint margin, which previously overflowed the screen by one line.
  • TUI scan view: wildcard and error notices are capped at the latest three (with a "+N earlier notices" line) and the viewport shrinks to fit them, so recursive scans no longer push the progress bar and status off-screen. Ctrl+C on a finished scan now quits instead of relabelling it "Aborted". The results buffer is capped at 10,000 lines and re-rendered on progress ticks once past 200 results, so large scans no longer slow the UI and back-pressure the workers. ~/ paths in the wordlist and output fields are expanded (#43).
  • -v lines no longer splice into the progress line. The dns package no longer writes to stderr; lookups log through scan.Config.Logf (the CLI passes output.Writer.Info), and Info/Error now take the writer mutex and blank the in-place progress line before printing (#36, ROADMAP N5).
  • tools/wordlist-gen output is deterministic: -combine no longer iterates a Go map, so the same flags give a byte-identical file. Write errors go to stderr and exit non-zero instead of being swallowed, suffixes such as .co.uk are stripped correctly, a prefix is no longer combined with itself, per-entry logging moved behind -v, and the generator has unit tests (#45).
  • The reliability-guard message printed as aborting scan: 50%!o(MISSING)f 200 queries failed: scan event messages were passed to Info/Error as format strings. They are now printed verbatim (present since 0.7.0).
  • Ctrl+C during the preflight lookup or wildcard probes no longer reports a misleading "resolver failed preflight" or "wildcard detection failed" error. A second Ctrl+C now force-quits a run that is stuck draining.
  • -o -- (a flag whose value is --) is no longer mistaken for the -- terminator.
  • On a terminal, results no longer start on the same line as the progress indicator, and a UTF-8 byte order mark at the start of a wordlist or -dL file no longer invalidates its first entry.

Added

  • -format jsonl streams one JSON object per resolved subdomain, so structured output can be piped live (#34).
  • -show-records appends TYPE=value record pairs to text output on stdout and in the -o file (#34, ROADMAP N2).
  • TUI form gains Max Names (-max-queries) and No Abort (-no-abort), both persisted in last.json. A recursive TUI scan above the 1e7 ceiling can now be started with a cap instead of enabling Force, which also disables the wildcard abort (#35).
  • -seed <n> makes simulation mode reproducible: each outcome, record and reported timing is derived from the seed and the name, so the same seed gives the same results. Without -seed a random seed is chosen and printed in the simulation banner (#44).
  • Tagged releases publish a multi-arch (linux/amd64, linux/arm64) image to ghcr.io/tmhsdigital/subenum with build provenance, tagged X.Y.Z, X.Y, and latest (non-prereleases only). The Dockerfile now cross-compiles in a build-platform builder stage, so no QEMU emulation is needed (#40).
  • -dL <file> scans every apex domain in a file (one per line, # comments allowed, - for stdin) in one process. Each domain is an independent scan with its own preflight, wildcard check, -max-queries budget and reliability guard; results share one output, so -format json stays a single array. A failing domain (for example a wildcard without -force) sets a non-zero exit code but does not stop the others. -w - reads the wordlist from stdin (#41, ROADMAP N4).

Changed

  • When stdout is not a terminal, text results are printed as bare subdomain names (no Found: prefix), so subenum ... | sort -u or | httpx works without cut. Terminal output is unchanged (#34).
  • Wordlists are normalized on load: blank lines and # comments are ignored, entries are lowercased before deduplication (so WWW and www are one query), and entries that are not valid DNS labels (whitespace, *, empty or over-long labels) or whose full name would exceed 253 characters are skipped and reported on stderr. Underscore labels such as _dmarc are kept. A wordlist with no valid entries is an error (#33).
  • -max-queries help text and docs now say what it counts: candidate names (jobs), not wire queries.
  • Release binaries and the Docker image are built with the latest stable Go toolchain (Docker builder golang:1.27.1-alpine, digest-pinned) instead of Go 1.24.2. go.mod still declares 1.24.2 as the minimum, and CI now tests both the minimum and the latest stable toolchain. CI runs govulncheck on every push and PR, and Dependabot tracks the Dockerfile base images. Release jobs build without a shared cache (#38).
  • -version no longer relies on a hand-maintained 0.7.0 fallback. Without ldflags it reports the module version recorded by go install (for example v0.7.0), or dev for local builds; the Makefile and Dockerfile fallbacks are dev too. README and the landing page document go install github.com/TMHSDigital/subenum@latest and add a "Why subenum" section (#39).
  • Simulation mode applies -hit-rate uniformly. Previously 12 common prefixes (www, api, dev, ...) always resolved about 90% of the time, so -hit-rate 1 still produced many hits and recursive simulate scans grew much faster than the rate suggested (#44).

Removed

  • Removed examples/multi_domain_scan.sh, which was bash-only; use -dL (#41).

0.7.0 - 2026-09-16

Added

  • Scan outcome accounting: each query is classified as found, nxdomain, timeout, refused, or other. The breakdown is printed after every scan (stderr) and carried on EventDone.Stats for the TUI summary line.
  • Reliability guard: once 200 queries have completed, if more than 20% failed as timeout/refused/other the scan emits an error naming the failure rate and likely resolver rate-limiting at the configured -t and -rate, then aborts. -no-abort keeps the warning but continues the scan.
  • TUI record-type selection (-type parity): a Record Types form field, persisted to the session config and wired into the scan.
  • TUI recursive enumeration (-recursive/-depth parity): a Recursive toggle with a Depth field gated on it, persisted across sessions.
  • TUI rate limiting (-rate parity): a queries-per-second form field, persisted across sessions.
  • TUI output file support (-o/-format parity): results can be written to a file as text, json, or csv. The format applies only to the file; the live viewport stays human-readable. Backed by a new file-only output.NewFile writer so structured output never collides with the alt-screen.
  • Release provenance attestations (actions/attest-build-provenance) on tagged builds.

Changed

  • dns.ResolveDomainWithRetry returns ([]Record, Outcome) instead of ([]Record, bool), so callers can distinguish NXDOMAIN from infrastructure failure.
  • ResolveDomainWithRetry no longer retries NXDOMAIN. Timeouts, refusals, and unknown errors still retry up to -attempts.
  • The resolver Dial hook honors network, so truncated UDP answers fall back to TCP. scan.Run builds one *net.Resolver and reuses it for every lookup.
  • Wildcard detection keeps a fingerprint of probe answers. Matching results are dropped (wildcardFiltered) even with -force. Recursive mode probes each new parent and skips expanding wildcard branches.
  • -max-queries caps admitted work. Recursive scans warn about the theoretical ceiling and refuse to start above 1e7 jobs unless -max-queries or -force is set. The dispatcher queue uses a head index with periodic compaction. Unreachable resolvers fail a preflight lookup before wildcard probes.
  • CLI EventError handling now drains the event channel so a reliability abort still delivers EventDone with partial stats. Wildcard-without--force still skips structured Finish.
  • TUI form now validates domain syntax and DNS server ip:port format up front, matching the CLI. The validators were extracted into a shared internal/validate package used by both entry points (previously the form only checked for non-empty values).
  • validate.Domain now enforces the 63-character per-label limit and accepts punycode TLDs (xn--...).
  • Version is a var injected via -ldflags "-X main.Version=$(git describe --tags --dirty)". -version writes to stdout.
  • go directive is 1.24.2 (required by charmbracelet/bubbles; a patchless go 1.24 makes go build demand tidy). golangci-lint CI is pinned to v2.12.2. Docker final stage is distroless static nonroot, with builder and runtime images pinned by digest. Tagged releases generate GitHub release notes.

Removed

  • Removed the unused dns.Resolve function, superseded by dns.ResolveTypes.
  • Collapsed the simulation helpers onto dns.SimulateResolve, removing the redundant SimulateResolution wrapper (its race-detection test was retargeted, not deleted).

Fixed

  • Structured output (-format json and -format csv) is now finalized only on the successful scan path. The finalizer previously ran via defer on every exit, so an early error (such as wildcard detection without -force) emitted an empty JSON array. Text output behavior is unchanged.
  • -version and the startup banner trim a leading v from the injected Version so git describe tags do not print as subenum vv0.7.0.

Docs

  • Rewrote the ARCHITECTURE data-flow section to describe the dispatcher-owned work queue instead of the removed feed-then-close subdomains channel model, and corrected the DNS engine function descriptions.
  • Refreshed the DEVELOPER_GUIDE: removed "Future Development" items that already shipped, updated the file tree, and corrected dns package references.
  • Fixed stale references in DOCUMENTATION_STRUCTURE.md (changelog path) and README (package blurbs).
  • Added docs/ROADMAP.md capturing the prioritized review findings and follow-up plan.
  • GitHub Pages landing page (docs/index.md) covers the 0.7.0 feature set: scan accounting, -no-abort, -max-queries, the recursion ceiling, and wildcard fingerprint filtering.
  • Normalized em dashes to hyphens across docs/ for consistency with the no-em-dash convention.
  • Documented outcome classification, EventDone.Stats, and the reliability guard in ARCHITECTURE.
  • Corrected ARCHITECTURE output-format note (TUI file output shipped in P5) and DEVELOPER_GUIDE "CLI-only zero deps" claim.

Tests

  • Added TUI coverage: session-config round-trip, form navigation across gated fields, record-type and depth validation, and structured-output finalization on both the success and error paths.
  • Added output-writer coverage: file-only writer stdout suppression, simulate-mode text prefix, and the CSV empty-record fallback row.
  • Moved the validator tests alongside the new internal/validate package.
  • Added resolver Classify tests and simulate-mode scan tests for outcome accounting, the reliability abort, and -no-abort.
  • Added a local UDP NXDOMAIN responder test proving -attempts 3 issues exactly one query for a definitive negative.
  • Added a local UDP+TCP responder test proving truncated (TC=1) answers complete over TCP.
  • Added wildcard fingerprint tests: matching answers are filtered, and recursive expansion of a wildcard branch is skipped.
  • Added tests for -max-queries admission, the recursive ceiling refusal, and resolver preflight against a black-holed UDP listener.
  • Replaced public-resolver tests with an in-process UDP/TCP responder. Live smoke is gated on SUBENUM_NETWORK_TESTS=1.

0.6.0 - 2026-06-03

Added

  • Resolved records are now captured during scans. internal/dns exposes Resolve and a Record{Type, Value} type; scan.Event carries Records for each resolved subdomain (A/AAAA today, extensible to CNAME and more).
  • -format text|json|csv flag (default text, byte-for-byte identical to prior output). JSON emits a buffered array of {"subdomain", "records"} objects; CSV streams subdomain,type,value rows with a header. The -o output file honors the selected format. Output formats are CLI-only for now (TUI-pending).
  • -rate <qps> flag (default 0 = unlimited) caps total DNS queries per second across the worker pool via a shared stdlib ticker gate inside scan.Run. The limiter respects context cancellation so Ctrl+C stays responsive.
  • -type A,AAAA,CNAME flag (default A,AAAA, preserving prior behavior) performs per-type DNS lookups and filters results to the requested types. The resolved record type is carried in the existing Record shape, so the JSON/CSV schema is unchanged.
  • -recursive and -depth <n> flags for recursive enumeration of discovered subdomains. scan.Run was restructured around a dispatcher that tracks outstanding work and closes the queue only when it drains to zero, so resolved subdomains can safely enqueue depth-capped children (the previous close-after-feed shape would have panicked on a send to a closed channel). A centralized visited set provides loop and duplicate protection, and the progress total expands as new work is discovered.

Changed

  • Internal: the scan engine's worker queue lifecycle moved from a feed-then-close channel to a dispatcher-owned queue with a pending-work counter.

0.5.1 - 2026-06-03

Fixed

  • Data race in simulation mode: migrated internal/dns/simulate.go to math/rand/v2, whose top-level functions are goroutine-safe and auto-seeded. SimulateResolution is now safe to call concurrently (previously a shared *math/rand.Rand was used from every worker).
  • Send-on-closed-channel race in scan.Run: the progress ticker goroutine now signals its own exit (tickerStopped) and guards its send with a select, and Run waits for that exit before emitting EventDone, so the deferred close(events) can no longer race an in-flight ticker send.
  • TUI now renders the "Aborted" status when a scan is cancelled with ctrl+c (the scan view is marked aborted so the subsequent EventDone shows partial counts).
  • TUI form no longer blocks a live-mode scan when the Hit Rate field is empty or out of range; hit rate is validated only when Simulate is on.
  • -version now reports the correct version (subenum v0.5.1).
  • Docker build: the builder now copies go.sum and runs go mod download before building, the base image satisfies the module Go version, and main_test.go is no longer copied into the build image.

Changed

  • Go minimum version reconciled to 1.24.2 across go.mod, the Dockerfile base image, README, and docs (the charmbracelet TUI dependencies require it); direct vs indirect dependency classification corrected via go mod tidy.

Added

  • Tests: concurrent SimulateResolution test, internal/scan runner tests (concurrent simulate run and mid-scan context cancellation), and internal/tui form validation tests.

Docs

  • README facelift: plain-text description under the badges, a copy-paste quick-start block, the TUI screenshot promoted to a hero position, PRs-Welcome and platform badges, and removal of em dashes for a clean human-authored look.
  • ARCHITECTURE: corrected the progress ticker interval (1 second) and the argument-parsing section (flag.*Var into a cliFlags struct).
  • Removed the unused, duplicate docs/assets/title.svg.

0.5.0 - 2026-03-14

Added

  • Interactive terminal UI (-tui flag) built with Bubble Tea — form-based config screen and live-scrolling results view; no CLI arguments required to launch
  • make tui Makefile target for one-command TUI launch
  • internal/scan package: extracted scan engine (scan.Run) with typed Event channel, usable by both CLI and TUI
  • TUI session persistence: last-used form values written to ~/.config/subenum/last.json and restored on next launch or after pressing r (new scan)

Changed

  • CLI scan loop in main.go now delegates to scan.Run() instead of containing the worker pool inline
  • External dependencies added: github.com/charmbracelet/bubbletea and github.com/charmbracelet/bubbles (TUI only; CLI path has zero external dependencies)
  • TUI form field order: Simulate toggle promoted to field 3 (was field 8); Hit Rate row is hidden when Simulate is OFF
  • TUI now shows a blinking cursor inside the active text input
  • Pressing r on the scan results screen returns to the form with last-used values pre-filled (was reset to defaults)

0.4.0 - 2026-03-14

Added

  • Wildcard DNS detection with double-probe confirmation; exits by default, continue with -force
  • Wordlist deduplication (duplicates removed before scanning, count reported in verbose mode)
  • -attempts flag replacing -retries (deprecated, still accepted with warning)

Changed

  • Refactored into internal/dns, internal/output, internal/wordlist packages
  • Progress, verbose, and diagnostic output moved to stderr (stdout is now pipe-clean)
  • Version bumped to 0.4.0

Fixed

  • Progress ticker no longer corrupts piped stdout output
  • -retries semantics clarified via rename to -attempts

[0.3.0] - 2026-02-22

Added

  • Output file support with the -o flag to save results to a file
  • DNS retry mechanism with configurable -retries flag for transient failure resilience
  • Graceful shutdown on SIGINT/SIGTERM — drains in-flight workers and prints partial results
  • Proper DNS server validation (IP format and port range 1-65535)
  • Domain format validation against DNS naming rules
  • Tests for validateDNSServer, validateDomain, resolveDomainWithRetry, and simulateResolution

Changed

  • Removed deprecated rand.Seed call (auto-seeded since Go 1.20)
  • Tests now use t.Errorf for real assertions instead of t.Logf warnings
  • Fixed test compilation — resolveDomain calls now pass all 4 required parameters
  • Updated all placeholder URLs/emails in documentation to actual repo values

Fixed

  • Progress goroutine done channel is now buffered to prevent potential deadlock
  • Mutex-protected stdout/file output to prevent interleaved writes from concurrent workers

[0.2.0] - 2025-05-08

Added

  • Custom DNS server support with the -dns-server flag
  • Verbose output mode with the -v flag
  • Progress reporting during scans (enabled by default)
  • Version information accessible via the -version flag
  • Input validation for concurrency and timeout values
  • Legal disclaimers and usage restrictions to prevent misuse
  • Comprehensive documentation via README.md and docs folder
  • Developer Guide with setup and contribution instructions
  • Example wordlist and multi-domain scanning script
  • Basic test suite for DNS resolution

Changed

  • Enhanced error handling with user-friendly messages
  • Improved code structure and documentation
  • DNS resolution now reports timing information in verbose mode

Fixed

  • Proper cleanup of resources after scans complete
  • Prevention of negative values for concurrency and timeout

[0.1.0] - 2025-05-07

Added

  • Initial project setup with basic functionality
  • Concurrent subdomain enumeration using goroutines
  • DNS resolution with configurable timeout
  • Command-line flags for wordlist, concurrency, and timeout