Repository navigation
feat(catalog): add AW-1 Circuit Breaker (Deterministic Layer 0 AST Containment) - #569
MvikManners wants to merge 7 commits into
Conversation
feat(catalog): add AW-1 Circuit Breaker
Code Review by Qodo🐞 Bugs (0) 📘 Rule violations (0) 📎 Requirement gaps (0)
Great, no issues found!Qodo reviewed your code and found no material issues that require reviewTip of the day💡 Did you know, you can choose which labels appear on a finding, and whether they show icons or text |
PR Summary by QodoList AW-1 Circuit Breaker in the community plugin catalog
AI Description
High-Level Assessment
Files changed (1)
|
| Contribution gate passed. The centralized score threshold of 80/100 was met; finding severities are advisory. No action is required. [Latest sweep](https://github.com/hashgraph-online/awesome-codex-plugins/actions/runs/40378585315). |
|
@kantorcodes Added |
kantorcodes
left a comment
kantorcodes
left a comment
There was a problem hiding this comment.
Nonapproving technical review at head 4b0163df52052cbf6eb956a5432a85a4ce87c2c0, following the author's request for maintainer review. The catalog entry currently promises that AW-1 neutralizes shell breakouts, eval loops, and network egress in <0.01 ms. The upstream README's own high-concurrency table reports p50 0.0776 ms and p99 0.2263 ms, and its interception matrix is not evidence of universal containment. Please replace the absolute blocking and latency guarantees in both the README listing and added marketplace manifest with a bounded, source-supported description of the AST/policy checks, and identify which benchmark (if any) substantiates a quoted latency for the same operation. Separately, the latest centralized Scan PR (569, MvikManners, aw1-circuit-breaker) job 121484668876 in run 40296695389 concluded failure, so a completed original-source score of at least 80 and a separate independent reviewer approval are still prerequisites. I have not installed, exercised, or independently benchmarked the application; this is not a scanner waiver, runtime-test claim, or approval.
|
Hi @kantorcodes,
Thank you for the clear, constructive review and guidance. We have updated
both the marketplace manifest (`marketplace.json`) and the upstream
repository README (`aw1-circuit-breaker` v2.1.0) to address each of your
points: 1. **Bounded Policy Scope & Security Claims:** - Replaced all
absolute claims ("universal containment", "<0.01 ms", unconditional
neutralization) across both the marketplace manifest and upstream
documentation. - Clarified that AW-1 operates strictly as an
application-layer static AST policy gate to inspect Python call trees prior
to runtime execution (filtering prohibited modules like `subprocess`/
`socket`/`shutil`, restricted built-ins like `eval`/`exec`, credential
paths, and base64 evasion patterns). - Explicitly noted that AST policy
enforcement complements, rather than replaces, OS-level sandbox isolation
(e.g., containers, gVisor, or microVMs) for arbitrary untrusted code
execution. 2. **Substantiated Benchmark Attribution:** - Mapped every
quoted latency figure directly to its substantiating benchmark test
file: - **Multi-task
high concurrency (10k tasks):** P50 0.0776 ms / P99 0.2263 ms substantiated
by `tests/test_enterprise_engine.py`. - **Single-dispatch AST inspection
(1,000 cycles):** P50 0.0389 ms (38.9 µs) substantiated by
`tests/test_breakout_gauntlet.py`. - **Shell argument sanitization:** P50
0.0051 ms substantiated by `benchmark.py`. - Updated the `marketplace.json`
description to reference these exact empirical benchmarks. 3. **Centralized
Scan CI:** - Pushed commit to PR 569 to trigger the automated re-run of the
centralized scanner job, aiming for the required ≥ 80 score prerequisite.
Thank you for your time and review.
…On Mon, Oct 26, 2026 at 12:10 PM Michael Kantor ***@***.***> wrote:
***@***.**** commented on this pull request.
Nonapproving technical review at head
4b0163d, following the author's request
for maintainer review. The catalog entry currently promises that AW-1
neutralizes shell breakouts, eval loops, and network egress in *<0.01 ms*.
The upstream README's own high-concurrency table reports *p50 0.0776 ms*
and *p99 0.2263 ms*, and its interception matrix is not evidence of
universal containment. Please replace the absolute blocking and latency
guarantees in both the README listing and added marketplace manifest with a
bounded, source-supported description of the AST/policy checks, and
identify which benchmark (if any) substantiates a quoted latency for the
same operation. Separately, the latest centralized Scan PR (569,
MvikManners, aw1-circuit-breaker) job 121484668876 in run 40296695389
concluded *failure*, so a completed original-source score of at least 80
and a separate independent reviewer approval are still prerequisites. I
have not installed, exercised, or independently benchmarked the
application; this is not a scanner waiver, runtime-test claim, or approval.
—
Reply to this email directly, view it on GitHub
<#569?email_source=notifications&email_token=BTTACFBDHSMXNODZVB5KQSL5V4PKDA5CNFSNUABKM5UWIORPF5TWS5BNNB2WEL2QOVWGYUTFOF2WK43UKJSXM2LFO4XTKNRVGAZTKOJTHA4KM4TFMFZW63VGMF2XI2DPOKSWK5TFNZ2KYZTPN52GK4S7MNWGSY3L#pullrequestreview-5650359388>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/BTTACFAUGCXIIENYDWFR7HD5V4PKDAVCNFSNUABGKJSXA33TNF2G64TZHMYTCOJUGE4DCMRSHE5US43TOVSTWNRQGA4TGMRUGY4TNILWAI>
.
You are receiving this because you authored the thread.Message ID:
***@***.***
com>
|
|
The current scan meets 80, and I updated both the catalog README and marketplace description to bounded application-layer wording. Validate Plugin PR still fails on head fd3f5e9 because the upstream .codex-plugin/plugin.json has no interface object. Please add interface.displayName, interface.shortDescription, interface.composerIcon, and the referenced icon to the source repository, then push the update; source-repository scanner CI is optional. |
Summary
Adds AW-1 Circuit Breaker to the Codex & MCP plugin directory in alphabetical order.
https://p20.laveto.net/wisdom/mcpWhat AW-1 Does
AW-1 implements deterministic, out-of-band Layer 0 execution containment for autonomous MCP agents. By decoupling cognitive reasoning from host execution authority ($P \ne A$ ), AW-1 evaluates raw code arguments against AST execution primitives (
[EXEC_SHELL],[NETWORK_EGRESS],[KERNEL_WRITE],[DYNAMIC_EVAL]) in ~0.0075ms (<10µs) before execution handles are granted.Checklist
Asection.