skills

Automatically cycles through implementation, PR creation, self-review, and fixes from an Issue or an instruction to produce a merge-ready PR. Supports single/multiple Issues/PRs and explicit dependency notation. Can optionally run through to merge. In orchestration, detects and self-resolves cross-cutting concerns before merge, leaving no follow-up within a single run.

Monitors Claude Code's Usage Limit (5-hour = Current / weekly = Weekly), pauses work when it exceeds 95%, and automatically resumes once the reset restores headroom. Bundles an engine form that a caller such as drive Reads at a checkpoint, and a standalone form, `/usage-guard "<continuation command>"`, that guards arbitrary work. Claude-only (depends on the OAuth usage-rate endpoint + ScheduleWakeup).

Monitors Claude Code's Usage Limit (5-hour = Current / weekly = Weekly), pauses work when it exceeds 95%, and automatically resumes once the reset restores headroom. Bundles an engine form that a caller such as drive Reads at a checkpoint, and a standalone form, `/usage-guard "<continuation command>"`, that guards arbitrary work. Claude-only (depends on the OAuth usage-rate endpoint + ScheduleWakeup).

Monitors Claude Code's Usage Limit (5-hour = Current / weekly = Weekly), pauses work when it exceeds 95%, and automatically resumes once the reset restores headroom. Bundles an engine form that a caller such as drive Reads at a checkpoint, and a standalone form, `/usage-guard "<continuation command>"`, that guards arbitrary work. Claude-only (depends on the OAuth usage-rate endpoint + ScheduleWakeup).

Collects open issues with the `backlog.mjs` engine, orders them by dependency graph (reusing drive's dependency-notation SSOT) and fixed-vocabulary priority rules, and presents start candidates output in drive argument format (e.g. `#12,#15 ->

A thin wrapper that collects logs from a failed CI run, formats the context, and connects to `/drive`. Input resolution (explicit run id > latest failure on an explicit branch > latest failure on the current branch) → flaky determination (one `gh run rerun --failed` + polling; skipped with `--no-rerun`) → log extraction (`gh run view --log-failed`; the ANSI-stripping + error-line extraction regex is the same as health's same-error determination) → assembling the instruction text → launching `/drive`. `--dry-run` outputs only the instruction text (no rerun, no drive launch). A failure on the main branch is called out as high priority at the top of the report.

Rules for generating Conventional Commits messages (Type/Scope decision table, format). Referenced by other skills.

Stages changes and commits them following Conventional Commits. Does not push or create a PR.

Enumerates automation PRs (renovate / dependabot etc.) with the `deps.mjs` engine and makes fixed-vocabulary judgments based on semver classification (from PR title / branch / manifest diff; grouped PRs use the maximum bump), CI status, lockfile consistency, and peer / engines changes. patch/minor + CI green + lockfile consistent → auto-merge candidate; major / CI red / pending / peer / engines → 要確認. Author judgment uses the same pattern as health area 15 (`*[bot]` / `app/*`; release-please is excluded — that's `/release`'s responsibility). `--dry-run` judges only; `--auto` executes without confirmation. Merging follows the central autonomy policy's irreversible gate (`--auto` overrides the policy).

Checks, in one shot via the `health-check.mjs` engine, for state unintentionally left behind during repository work (working tree, stash, branch, worktree, PR, issue, actions, etc.) and skill-catalog consistency, presenting a 16-area status table and fixed-vocabulary recommended actions. With `--deep`, does additional read-only investigation of `要確認` items; with `--fix`, executes safe-vocabulary actions (prune / delete / fetch, and `drop` when promoted by `--deep`) following the central policy's gate (reversible-local=proceed / irreversible=ask). Read-only by default.

Creates a branch, plans the implementation, and makes code changes based on an Issue or an instruction. Accepts an Issue number or a text instruction.

Consumes the session lessons queue (~/.agents/lessons/queue.jsonl), extracts lessons about User Skills from the transcript, and files only the approved ones as issues to ozzy-labs/skills. Fires on "organize the lessons," "consume the lessons," "session retrospective."

Generates a Phase-N tracking issue. Assembles a structured issue body containing cross-session handoff context, a decisions table, per-PR tasks, DoD, and the Phase N+1 outlook, then files it with gh issue create. Supports a non-interactive mode where all items are passed as arguments, and an interactive mode (Claude Code companion) that fills in whatever is missing.

Defines the contract (policy.schema.json) and read substrate (policy-read.mjs) for the central autonomy policy. The SSOT for the 3 action classes, gate vocabulary, policy.yaml hierarchy, and zero-config defaults. Referenced by other skills and hooks.

Pushes committed changes to the remote and creates or updates a PR.

Detects release-please PRs (`gh pr list --author app/release-please --state open`), validates them against a fixed checklist (SemVer consistency between version bump and included commit types — feat→minor / fix→patch / `!` or BREAKING CHANGE→major —, CHANGELOG consistency, CI all green), then merges via `gh pr merge --squash` through an approval gate by default (externally visible / effectively irreversible). After merging, it monitors the publish workflow via polling (30s interval, 20-minute cap) and confirms rollout with `npm view <pkg> version` (npm-distributed repos only). `--auto` skips the gate only when all validations pass (stops if any fail). npm publish assumes OIDC Trusted Publishers (no `NPM_TOKEN` used).

Reviews code changes or PRs across 11 perspectives, reporting via JSON structured output plus a human-readable report. Supports switching between quick and deep modes. Takes a PR number or working-tree diff as input.

Runs verification, commit, and PR creation all at once. An integrated pipeline that runs verify (build + type + test + lint) → commit → PR creation in sequence against the changes.

Aggregates the local observability event log (~/.agents/observability/events.jsonl) and presents, read-only, per-skill invocation counts and notable events (fallback / HITL rejection / loop cap reached / abort). Fires on "show me the skill metrics," "aggregate the observability results," "which skills get used the most?" Does not send anything.

Defines the skill observability event contract (event.schema.json) and emit substrate (obs-emit.mjs). The SSOT for the measurement layer of the skill improvement loop. Referenced by other skills and hooks.

Selects GitHub topic candidates by having the `topics.mjs` engine perform constraint validation, popularity measurement, broad+narrow / singular-plural comparison, and hardcoded ozzy-labs conventions, then applies them via `gh repo edit --add-topic` following the policy's `externally-visible` gate (default batch-confirm). Scope is limited to use within ozzy-labs.

An integrated skill that runs the combined validation of build / typecheck / test / lint in one shot. The `verify.mjs` engine auto-discovers validation commands via a discovery chain (AGENTS.md's 「検証」 section → package.json scripts → justfile/Makefile/lefthook → language heuristics), executes them serially with their source attached, and returns a result summary. If a higher-priority stage produces a hit, only that stage is executed (no crossing stages).

Defines the skill observability event contract (event.schema.json) and emit substrate (obs-emit.mjs). The SSOT for the measurement layer of the skill improvement loop. Referenced by other skills and hooks.

Collects open issues with the `backlog.mjs` engine, orders them by dependency graph (reusing drive's dependency-notation SSOT) and fixed-vocabulary priority rules, and presents start candidates output in drive argument format (e.g. `#12,#15 ->

A thin wrapper that collects logs from a failed CI run, formats the context, and connects to `/drive`. Input resolution (explicit run id > latest failure on an explicit branch > latest failure on the current branch) → flaky determination (one `gh run rerun --failed` + polling; skipped with `--no-rerun`) → log extraction (`gh run view --log-failed`; the ANSI-stripping + error-line extraction regex is the same as health's same-error determination) → assembling the instruction text → launching `/drive`. `--dry-run` outputs only the instruction text (no rerun, no drive launch). A failure on the main branch is called out as high priority at the top of the report.

Rules for generating Conventional Commits messages (Type/Scope decision table, format). Referenced by other skills.

Stages changes and commits them following Conventional Commits. Does not push or create a PR.

Enumerates automation PRs (renovate / dependabot etc.) with the `deps.mjs` engine and makes fixed-vocabulary judgments based on semver classification (from PR title / branch / manifest diff; grouped PRs use the maximum bump), CI status, lockfile consistency, and peer / engines changes. patch/minor + CI green + lockfile consistent → auto-merge candidate; major / CI red / pending / peer / engines → 要確認. Author judgment uses the same pattern as health area 15 (`*[bot]` / `app/*`; release-please is excluded — that's `/release`'s responsibility). `--dry-run` judges only; `--auto` executes without confirmation. Merging follows the central autonomy policy's irreversible gate (`--auto` overrides the policy).

Automatically cycles through implementation, PR creation, self-review, and fixes from an Issue or an instruction to produce a merge-ready PR. Supports single/multiple Issues/PRs and explicit dependency notation. Can optionally run through to merge. In orchestration, detects and self-resolves cross-cutting concerns before merge, leaving no follow-up within a single run.

Creates a branch, plans the implementation, and makes code changes based on an Issue or an instruction. Accepts an Issue number or a text instruction.

Consumes the session lessons queue (~/.agents/lessons/queue.jsonl), extracts lessons about User Skills from the transcript, and files only the approved ones as issues to ozzy-labs/skills. Fires on "organize the lessons," "consume the lessons," "session retrospective."

Defines the contract (policy.schema.json) and read substrate (policy-read.mjs) for the central autonomy policy. The SSOT for the 3 action classes, gate vocabulary, policy.yaml hierarchy, and zero-config defaults. Referenced by other skills and hooks.

Pushes committed changes to the remote and creates or updates a PR.

Detects release-please PRs (`gh pr list --author app/release-please --state open`), validates them against a fixed checklist (SemVer consistency between version bump and included commit types — feat→minor / fix→patch / `!` or BREAKING CHANGE→major —, CHANGELOG consistency, CI all green), then merges via `gh pr merge --squash` through an approval gate by default (externally visible / effectively irreversible). After merging, it monitors the publish workflow via polling (30s interval, 20-minute cap) and confirms rollout with `npm view <pkg> version` (npm-distributed repos only). `--auto` skips the gate only when all validations pass (stops if any fail). npm publish assumes OIDC Trusted Publishers (no `NPM_TOKEN` used).

Reviews code changes or PRs across 11 perspectives, reporting via JSON structured output plus a human-readable report. Supports switching between quick and deep modes. Takes a PR number or working-tree diff as input.

Runs verification, commit, and PR creation all at once. An integrated pipeline that runs verify (build + type + test + lint) → commit → PR creation in sequence against the changes.

Aggregates the local observability event log (~/.agents/observability/events.jsonl) and presents, read-only, per-skill invocation counts and notable events (fallback / HITL rejection / loop cap reached / abort). Fires on "show me the skill metrics," "aggregate the observability results," "which skills get used the most?" Does not send anything.

An integrated skill that runs the combined validation of build / typecheck / test / lint in one shot. The `verify.mjs` engine auto-discovers validation commands via a discovery chain (AGENTS.md's 「検証」 section → package.json scripts → justfile/Makefile/lefthook → language heuristics), executes them serially with their source attached, and returns a result summary. If a higher-priority stage produces a hit, only that stage is executed (no crossing stages).

Collects open issues with the `backlog.mjs` engine, orders them by dependency graph (reusing drive's dependency-notation SSOT) and fixed-vocabulary priority rules, and presents start candidates output in drive argument format (e.g. `#12,#15 ->

A thin wrapper that collects logs from a failed CI run, formats the context, and connects to `/drive`. Input resolution (explicit run id > latest failure on an explicit branch > latest failure on the current branch) → flaky determination (one `gh run rerun --failed` + polling; skipped with `--no-rerun`) → log extraction (`gh run view --log-failed`; the ANSI-stripping + error-line extraction regex is the same as health's same-error determination) → assembling the instruction text → launching `/drive`. `--dry-run` outputs only the instruction text (no rerun, no drive launch). A failure on the main branch is called out as high priority at the top of the report.