ryl

Use when cutting a ryl release. Covers the lockstep version bump across the five version-bearing files, the lockfile refresh, the tag/push gate, and the post-release SchemaStore + Trusted-Publishing flow.

Use when a ryl winget-pkgs validation PR is blocked by a Microsoft Defender false positive (`Validation-Defender-Error` / "Installer failed security check"). Covers the transient-vs-reproducible decision, local Defender + VirusTotal reproduction, the WDSI false-positive submission, and the log noise to ignore.

Lint and auto-fix YAML with the ryl CLI (a fast, yamllint-compatible linter). Use when running or configuring ryl in a repo, fixing YAML lint errors, setting up YAML linting, or migrating from yamllint. Covers the no-default-on-rules gotcha (ryl exits 2 without a config), exit codes, --fix vs --diff, machine-readable output, the YAML-vs-TOML config split, inline directives, and YAML embedded in Markdown.

Use when filing or editing a GitHub issue/PR for ryl or, especially, for another project (yamllint, granit, SchemaStore). Covers the cross-repo reference footgun, the draft-locally-then-file gate for other people's repos, the never-on-an-assumption rule, and the verified-Sources requirement — so a published issue does not need a verify-then-re-push cycle.

Use when adding or changing a rule's detection or safe-fix behaviour, or editing any property-test suite (safe-fix / rule-checker / markdown-fix / config). Covers what each generator must be extended with, the ~1000x pre-commit run, and which rules intentionally have no safe `--fix`.

Drive a Codex CI code review on a GitHub PR end to end, from triggering it through monitoring and classifying the verdict (clean / findings / rate-limited) to handling the resulting comments (react, resolve, reply). Use after pushing changes you want Codex to (re-)review, while waiting on a verdict, or when triaging Codex's review comments. It encodes the polling gotchas (REST not GraphQL, three verdict channels, transient 👀) and the comment-handling conventions (when to thumbs-up/down, resolve, or reply).

Use when chasing missed lines/regions to satisfy CI's zero-missed-coverage gate, or writing tests to close uncovered ranges. Covers the coverage-missing.py workflow, coverage-friendly Rust idioms, and the common uncovered hotspots.

Use to run a development-friction retrospective over recent Claude Code sessions: a cheap deterministic extractor digests the transcripts (tool errors, wall-clock waits, retry clusters, corrections), then a small number of classifier agents tag the digests against a fixed taxonomy. Prefer this over fanning a classifier agent at each raw transcript (far cheaper, less noise).

Use when adding a new ryl lint rule or changing an existing rule's wiring. Walks the multi-site checklist (rule module, registration, dispatch, TOML config, tests, docs) where a missed site usually trips a guard test rather than shipping silently.