Skip to content

Commit 4907144

Browse files
committed
fix(agent): align worker tool names and skills with the consolidated fleet
The worker prompt listed write, edit and delete on gpt while the wire carries apply_patch; it now lists what the wire carries. Skills that told agents to spawn builder, critic, intern or greybeard point at coder, reviewer and planner, and stale director names in comments are gone.
1 parent 9a5b9f9 commit 4907144

12 files changed

Lines changed: 54 additions & 46 deletions

File tree

‎evals/capability/README.md‎

Lines changed: 16 additions & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -158,7 +158,7 @@ bun run eval:capability -- --provider <name> --model <id> --repeats 5 \
158158
--baseline evals/capability/results/baseline-0286.json
159159

160160
# Overlay a closed-fleet director on the product exec path (eval/CI override,
161-
# not single-agent mode). Omit / skywalker keep the default Skywalker session.
161+
# not single-agent mode). Omit / dispatch keep the default Dispatch session.
162162
bun run eval:capability -- --provider <name> --model <id> --director build
163163
```
164164

@@ -183,21 +183,21 @@ change that justifies it.
183183

184184
Flags:
185185

186-
| Flag | Meaning |
187-
| ------------------------------------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
188-
| `--case <id\|all>` | Case id or `all` (default) |
189-
| `--provider <name>` / `--model <id>` | Required unless `--matrix`. Single-variant via `loadConfig`. Not inferred from local settings |
190-
| `--matrix <cells>` | Alternative to `--provider`/`--model`. Multi-variant: `p:m,p2:m2` or `label=p:m` (comma-separated). Every cell must include both sides |
191-
| `--config <path>` | Settings file override (CI injection) |
192-
| `--out <path>` | Write machine-readable results JSON |
193-
| `--baseline <path>` | Compare this run to a prior results file (improve/regress + metric deltas) |
194-
| `--ask-permissions` | Do **not** pass `--dangerously-skip-permissions` |
195-
| `--agent-timeout-ms <n>` | Wall-clock limit for `runExec` (default `1200000`, env `CORBITS_EVAL_AGENT_TIMEOUT_MS`) |
196-
| `--verify-timeout-ms <n>` | Wall-clock limit for `verify.sh` (default `120000`, env `CORBITS_EVAL_VERIFY_TIMEOUT_MS`) |
197-
| `--repeats <n>` | Runs per case×variant cell (default `1`; gate runs use `5`, baseline freezes `3`). Results record every repeat plus per-cell aggregates |
198-
| `--concurrency <n>` | Independent case×variant×repeat cells in parallel (default `1`, env `CORBITS_EVAL_CONCURRENCY`). Each cell still uses its own temp workdir. Use `--concurrency 4` (or similar) to run a live matrix faster |
199-
| `--dry-run` | Load cases × variants and print plan; no inference. Still requires `--provider`/`--model` or `--matrix` |
200-
| `--director <id>` | Exec overlay: run the product `corbits exec` path with this director's system prompt and initially-advertised tool set (default: skywalker). Eval/CI override, not single-agent mode. Directors that cannot spawn (for example `build`) do not mount `spawn_agent` / `wait_agents`. |
186+
| Flag | Meaning |
187+
| ------------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
188+
| `--case <id\|all>` | Case id or `all` (default) |
189+
| `--provider <name>` / `--model <id>` | Required unless `--matrix`. Single-variant via `loadConfig`. Not inferred from local settings |
190+
| `--matrix <cells>` | Alternative to `--provider`/`--model`. Multi-variant: `p:m,p2:m2` or `label=p:m` (comma-separated). Every cell must include both sides |
191+
| `--config <path>` | Settings file override (CI injection) |
192+
| `--out <path>` | Write machine-readable results JSON |
193+
| `--baseline <path>` | Compare this run to a prior results file (improve/regress + metric deltas) |
194+
| `--ask-permissions` | Do **not** pass `--dangerously-skip-permissions` |
195+
| `--agent-timeout-ms <n>` | Wall-clock limit for `runExec` (default `1200000`, env `CORBITS_EVAL_AGENT_TIMEOUT_MS`) |
196+
| `--verify-timeout-ms <n>` | Wall-clock limit for `verify.sh` (default `120000`, env `CORBITS_EVAL_VERIFY_TIMEOUT_MS`) |
197+
| `--repeats <n>` | Runs per case×variant cell (default `1`; gate runs use `5`, baseline freezes `3`). Results record every repeat plus per-cell aggregates |
198+
| `--concurrency <n>` | Independent case×variant×repeat cells in parallel (default `1`, env `CORBITS_EVAL_CONCURRENCY`). Each cell still uses its own temp workdir. Use `--concurrency 4` (or similar) to run a live matrix faster |
199+
| `--dry-run` | Load cases × variants and print plan; no inference. Still requires `--provider`/`--model` or `--matrix` |
200+
| `--director <id>` | Exec overlay: run the product `corbits exec` path with this director's system prompt and initially-advertised tool set (default: dispatch). Eval/CI override, not single-agent mode. Directors that cannot spawn (for example `build`) do not mount `spawn_agent` / `wait_agents`. |
201201

202202
## Case format
203203

‎plugins/corbits-skills/skills/git-worktrees/SKILL.md‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -7,7 +7,7 @@ description: Create a git worktree from origin/<default-branch> and tear it down
77

88
# git-worktrees
99

10-
How to create a worktree from origin/<default-branch> and tear it down. Load via `use_skill("git-worktrees")` and copy commands into an intern brief. Intern executes via `run_shell`. Do not run the git on the parent.
10+
How to create a worktree from origin/<default-branch> and tear it down. Load via `use_skill("git-worktrees")` and copy commands into a coder brief. The coder executes via `run_shell`. Do not run the git on the parent.
1111

1212
## Create from origin/<default-branch>
1313

@@ -17,7 +17,7 @@ git fetch origin
1717
git worktree add ../worktree/<branch-name> -b <branch-name> origin/<default-branch>
1818
```
1919

20-
Always base new branches on `origin/<default-branch>` (whatever the repository uses). After creating the worktree, intern `cd`s into it and installs local dependencies (`bun install` when the project uses Bun; otherwise follow developer docs). Worktrees do not share `node_modules`.
20+
Always base new branches on `origin/<default-branch>` (whatever the repository uses). After creating the worktree, the coder `cd`s into it and installs local dependencies (`bun install` when the project uses Bun; otherwise follow developer docs). Worktrees do not share `node_modules`.
2121

2222
## Teardown
2323

‎plugins/corbits-skills/skills/lexicon/SKILL.md‎

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -39,16 +39,16 @@ checkout — never float mid-run.
3939
`src/agent/directors/<id>/package.ts` as `systemPrompt`.
4040
- Agents side: `plugins/*/agents/*.md` files in the checkout, matched
4141
by file basename — `<id>.md` matches director `<id>` exactly.
42-
- Near-misses are not diffs: `critique.md` is not `critic`, and
43-
`marketing-intern.md` is not `intern`. List them as unmatched, do
42+
- Near-misses are not diffs: `critique.md` is not `reviewer`, and
43+
`marketing-intern.md` is not `coder`. List them as unmatched, do
4444
not force a comparison.
4545

4646
## Step 3: Diff same-name pairs
4747

4848
For each matched pair, compare the director `systemPrompt` against the
4949
agents file body at the pin. Report per director: in sync, or drifted
5050
with the drifted sections quoted on both sides. Note the ported-from
51-
commit recorded in the package comment (e.g. gaasbot's `@ 6e16b6c`)
51+
commit recorded in the package comment (e.g. `@ 6e16b6c`)
5252
when it disagrees with the pin — a stale port marker is itself drift.
5353

5454
## Step 4: Report assembled prompt sizes

‎plugins/corbits-skills/skills/native-integration/SKILL.md‎

Lines changed: 13 additions & 13 deletions
Original file line numberDiff line numberDiff line change
@@ -28,9 +28,9 @@ When a GaaS skill names a Claude/GaaS tool, use the Corbits equivalent. Do not c
2828
| TaskUpdate | `manage_tasks` |
2929
| TaskList | `manage_tasks` |
3030
| AskUserQuestion | `ask_operator` (primary) / `ask_director` (worker) |
31-
| `Task` / `@greybeard` | `spawn_agent(agent="greybeard")` |
32-
| `@critic` / `@critique` | `spawn_agent(agent="critic")` |
33-
| `@intern` | `spawn_agent(agent="intern")` |
31+
| `Task` / `@greybeard` | `spawn_agent(agent="planner")` |
32+
| `@critic` / `@critique` | `spawn_agent(agent="reviewer")` |
33+
| `@intern` | `spawn_agent(agent="coder")` |
3434
| `@explorer` | `spawn_agent(agent="explorer")` |
3535
| Read / Write / Edit | `read` / `write` / `edit` |
3636
| Glob / Grep | `glob` / `grep` |
@@ -49,9 +49,9 @@ GaaS refactor says "ask clarifying questions" / "ask the user". Corbits extras:
4949

5050
GaaS scribe uses the `question` tool. Corbits extras: `ask_operator` (tool mapping above). Do not fork the GaaS scribe body.
5151

52-
When GaaS implement says you are orchestrated by karen, that is the Corbits primary (Skywalker). Route those disposition decisions through the primary, not a worker.
52+
When GaaS implement says you are orchestrated by karen, that is the Corbits primary (Dispatch). Route those disposition decisions through the primary, not a worker.
5353

54-
GaaS implement "Initial Planning" / Greybeard-before-code is not `/plan`. Substantial Builder work consumes a counsel / `/plan` plan (files, acceptance criteria, non-goals, risks, ordered steps) and blocks if that plan is missing. Tiny parent-DIY stays plan-optional. `/plan` and counsel author; they do not ship. `/implement` does not steal planning from `/plan`. Do not fork the GaaS implement body.
54+
GaaS implement "Initial Planning" / Greybeard-before-code is not `/plan`. Substantial coder work consumes a planner / `/plan` plan (files, acceptance criteria, non-goals, risks, ordered steps) and blocks if that plan is missing. Tiny parent-DIY stays plan-optional. `/plan` and planner author; they do not ship. `/implement` does not steal planning from `/plan`. Do not fork the GaaS implement body.
5555

5656
## Linear claim-first
5757

@@ -67,7 +67,7 @@ GaaS linear-issue-workflow inlines `git worktree add` and marks In Progress afte
6767

6868
GaaS `style` refuses to operate outside a git repo. Corbits does not: a folder without `.git` is a valid working directory (scratch, unpacked tarball, new project). Git-using skills (`implement`, `review`, `git-rebase`, `pull-request-review`) still no-op or ask when they need a repo. Do not invent a git repo to satisfy those skills.
6969

70-
When GaaS git-rebase writes `/tmp` editor scripts, Corbits still plans on the primary and intern executes sequenced git via `bash`; intern may use inline `GIT_SEQUENCE_EDITOR` instead of write editor scripts. Do not fork the GaaS git-rebase body.
70+
When GaaS git-rebase writes `/tmp` editor scripts, Corbits still plans on the primary and the coder executes sequenced git via `bash`; it may use inline `GIT_SEQUENCE_EDITOR` instead of write editor scripts. Do not fork the GaaS git-rebase body.
7171

7272
## Tracker-agnostic issues
7373

@@ -100,14 +100,14 @@ Do **not** post when the user only asked for a private/local read with no PR, or
100100

101101
### Multi-persona reviews
102102

103-
When the workflow ran more than one review lens (for example `critic` for behavioral/architecture, `greybeard` for waivers or product judgment, an OSS/quality agent for packaging and public-API bar), each lens that produced a distinct judgment **posts its own review**. Do not collapse independent verdicts into one mushy "team thinks" paragraph.
103+
When the workflow ran more than one review lens (for example `reviewer` for behavioral/architecture, `planner` for waivers or product judgment, an OSS/quality agent for packaging and public-API bar), each lens that produced a distinct judgment **posts its own review**. Do not collapse independent verdicts into one mushy "team thinks" paragraph.
104104

105-
| Lens | What it owns | When to post |
106-
| ---------------------- | ------------------------------------------------------------------ | ----------------------------------------------------------------------------------------- |
107-
| Primary / orchestrator | Verdict on the branch as it stands; residual findings; waiver list | Always when posting |
108-
| Critic | Behavioral bugs, missing tests, architecture, commit coherence | When a critic director ran |
109-
| Greybeard | Waiver rulings and intentional exceptions | When Greybeard authorized any waiver, or when product/architecture judgment was requested |
110-
| OSS / quality | Public-API, packaging, polish bar for shippable surface | When that lens was explicitly run |
105+
| Lens | What it owns | When to post |
106+
| ---------------------- | ------------------------------------------------------------------ | ------------------------------------------------------------------------------------------- |
107+
| Primary / orchestrator | Verdict on the branch as it stands; residual findings; waiver list | Always when posting |
108+
| Reviewer | Behavioral bugs, missing tests, architecture, commit coherence | When a reviewer director ran |
109+
| Planner | Waiver rulings and intentional exceptions | When the planner authorized any waiver, or when product/architecture judgment was requested |
110+
| OSS / quality | Public-API, packaging, polish bar for shippable surface | When that lens was explicitly run |
111111

112112
Same GitHub account is fine. Label each post so a human can tell which lens spoke. Prefer separate `gh pr review` / `gh pr comment` posts over one mega-comment when more than one lens has substance.
113113

‎plugins/corbits-skills/skills/ponytail/SKILL.md‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,7 @@
11
---
22
name: ponytail
33
user-invocable: false
4-
description: Compact builder mode guidance for minimal safe implementation diffs.
4+
description: Compact coder mode guidance for minimal safe implementation diffs.
55
---
66

77
Ponytail is a Builder discipline for keeping implementation diffs small without
@@ -29,5 +29,5 @@ accessibility, data integrity, tests, repo conventions, operator requirements,
2929
and explicit success criteria outrank minimal LOC. A mode never weakens those
3030
constraints.
3131

32-
For review or audit, treat Ponytail as a lens for critic, neckbeard, or primary
32+
For review or audit, treat Ponytail as a lens for reviewer or primary
3333
instructions; do not create a Ponytail director or agent.

‎src/agent/directors/shakespeare/package.ts‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -13,7 +13,7 @@ export const shakespearePackage: DirectorPackage = {
1313
"shipping product features",
1414
"pure code review",
1515
"orchestration / fleet control",
16-
"acting as tester or implementer",
16+
"acting as reviewer or implementer",
1717
],
1818
description: "Docs maintenance — PRODUCT / ARCHITECTURE / IMPLEMENTATION",
1919
systemPrompt: `You are ShakespeareDirector (Shakespeare), a specialist in Corbits Code.

0 commit comments

Comments
 (0)