Regenerate the core module index and document the modules the gate work touched #503

Merged

mph merged 1 commit from chore/regenerate-skills-0.11.0 into main

2026-09-02 21:21:08 +00:00

mph commented

2026-09-02 21:08:30 +00:00

Owner

Regenerates the module index and the API pages for the modules PR #501 touched, ahead of the v0.11.0 release. Refs #440.

llms.txt

It listed 26 of the 120 modules under core/model, core/adapters and core/output, and three of those 26 entries — ServerTypes, WorkflowTypes, WorktreeTypes — are pages, not modules. It now carries one entry per real module, 120 in all, each description taken from that file's own PURPOSE header.

Linking follows #440's stated rule that a link to a page that does not exist is worse than no link:

Form	Count
Linked to its own page	30
Linked to the aggregate page that documents it	12
Named and described, no link	78

The twelve are the modules covered inside ServerTypes.md, WorkflowTypes.md and WorktreeTypes.md — those pages answer for them, so the entry points there rather than at a page that was never written. No dangling link is introduced.

docs/api

Seven modules gain a page — all of them changed by #501 and none previously documented:

BatchImplement (including the new AwaitCI outcome and the poll-don't-re-dispatch rule), CiGate (the verdict model, the config defaults, and the supersession rule), PhaseArgs (including repeatedArg's silent drop of a value beginning with two dashes), ReviewStateCliParser, ReviewStateUpdater, ReviewStateValidator, ScalafixGate.

Git.md is updated in place for push now carrying the hook's report on failure.

Not in this PR

The remaining ~78 undocumented modules stay with #440. Note for that issue: it estimates "roughly 22 modules"; the real figure is about 78 after accounting for the three aggregate pages. It understates its own scope by more than 3×.

Published skills

build/published-skills/ is gitignored, so the two SKILL.md files are not in this diff. They were regenerated in the same pass and are staged for copying into the dev-docs plugin. Regenerating them surfaced several errors in the published April copy, all now corrected:

It documented implement and batch-implement as commands. Neither exists in commands/.
Exit code 3 was undocumented anywhere — the "wait unfinished, run me again" contract for phase-pr, phase-merge and ci-gate. This is the single most important thing a calling agent needs and it existed only in the kanon skill text.
./iw config get tracker and team-prefix are not valid fields.
Forgejo was missing from the tracker list, and FORGEJO_API_TOKEN from the environment section.
It claimed bare ./iw lists commands; it prints a usage banner (./iw --list lists them).
Effort worktrees, ci-gate, gate, doctor --env|--quality|--fix, rm --path|--no-prompt|--expect-dirty and dashboard --dev|--state-path were all absent.

🤖 Generated with Claude Code

https://claude.ai/code/session_013UpRzj7oiV1UDArn4hfu7U

Regenerates the module index and the API pages for the modules PR #501 touched, ahead of the v0.11.0 release. Refs #440. ## llms.txt It listed 26 of the 120 modules under `core/model`, `core/adapters` and `core/output`, and three of those 26 entries — `ServerTypes`, `WorkflowTypes`, `WorktreeTypes` — are pages, not modules. It now carries one entry per real module, 120 in all, each description taken from that file's own `PURPOSE` header. Linking follows #440's stated rule that a link to a page that does not exist is worse than no link: | Form | Count | |---|---| | Linked to its own page | 30 | | Linked to the aggregate page that documents it | 12 | | Named and described, no link | 78 | The twelve are the modules covered inside `ServerTypes.md`, `WorkflowTypes.md` and `WorktreeTypes.md` — those pages answer for them, so the entry points there rather than at a page that was never written. **No dangling link is introduced.** ## docs/api Seven modules gain a page — all of them changed by #501 and none previously documented: `BatchImplement` (including the new `AwaitCI` outcome and the poll-don't-re-dispatch rule), `CiGate` (the verdict model, the config defaults, and the supersession rule), `PhaseArgs` (including `repeatedArg`'s silent drop of a value beginning with two dashes), `ReviewStateCliParser`, `ReviewStateUpdater`, `ReviewStateValidator`, `ScalafixGate`. `Git.md` is updated in place for `push` now carrying the hook's report on failure. ## Not in this PR The remaining ~78 undocumented modules stay with #440. Note for that issue: it estimates "roughly 22 modules"; the real figure is about 78 after accounting for the three aggregate pages. It understates its own scope by more than 3×. ## Published skills `build/published-skills/` is gitignored, so the two `SKILL.md` files are not in this diff. They were regenerated in the same pass and are staged for copying into the dev-docs plugin. Regenerating them surfaced several errors in the published April copy, all now corrected: - It documented `implement` and `batch-implement` as commands. **Neither exists** in `commands/`. - **Exit code 3 was undocumented anywhere** — the "wait unfinished, run me again" contract for `phase-pr`, `phase-merge` and `ci-gate`. This is the single most important thing a calling agent needs and it existed only in the kanon skill text. - `./iw config get tracker` and `team-prefix` are not valid fields. - Forgejo was missing from the tracker list, and `FORGEJO_API_TOKEN` from the environment section. - It claimed bare `./iw` lists commands; it prints a usage banner (`./iw --list` lists them). - Effort worktrees, `ci-gate`, `gate`, `doctor --env|--quality|--fix`, `rm --path|--no-prompt|--expect-dirty` and `dashboard --dev|--state-path` were all absent. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_013UpRzj7oiV1UDArn4hfu7U

mph added 1 commit

2026-09-02 21:08:30 +00:00

docs(api): index every core module and document the modules the gate work touched

CI / contract (pull_request) Has been skipped

Details

CI / format (pull_request) Successful in 25s

Details

CI / lint (pull_request) Successful in 8s

Details

CI / compile (pull_request) Successful in 1m47s

Details

CI / dashboard-build (pull_request) Successful in 1m15s

Details

CI / test (pull_request) Successful in 10m30s

Details

6ffa8e246a

llms.txt listed three aggregate names that exist as pages but not as modules,
and named 26 of the 120 modules under core/model, core/adapters and core/output.
It now carries one entry per module, each description taken from that file's
PURPOSE header. A module is linked when a page can answer for it: its own page
where one exists, and the aggregate page that documents it for the twelve
covered by ServerTypes, WorkflowTypes and WorktreeTypes. The remaining 78 are
named and described without a link, because a link to a page that does not
exist is worse than no link.

Seven modules gain a page: BatchImplement, CiGate, PhaseArgs,
ReviewStateCliParser, ReviewStateUpdater, ReviewStateValidator and ScalafixGate.
Git's page gains the push failure that carries the hook's report.

Refs: #440

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013UpRzj7oiV1UDArn4hfu7U

mph merged commit 4529c4882d into main

2026-09-02 21:21:08 +00:00

mph referenced this pull request from a commit

2026-09-02 21:21:09 +00:00

Merge pull request 'Regenerate the core module index and document the modules the gate work touched' (#503) from chore/regenerate-skills-0.11.0 into main