Use to mine ../airlineops for governance habits gzkit lacks; produces a parity scan report (gzkit leads, AirlineOps adopts).
日本語の概要は準備中です。原文の説明を表示しています。
Use after a skill or governance-surface change to regenerate the control surfaces and skill mirrors.
インストールする前に、エージェントに与えられる指示の中身を確認できます。
gz agent sync control-surfaces (_run_agent_control_sync in
src/gzkit/commands/tidy.py) regenerates every derived control surface from
canon through sync_all in src/gzkit/sync_surfaces.py:
paths.manifest, and the Codex baseline .codex/config.toml
when it is missing or empty;.gzkit.json declares no vendors, .codex/hooks.json
and .codex/agents/*.toml (see § Codex delivery);src/gzkit/<surface>/
for skills, rules, personas, templates and canonical-class chores;AGENTS.md, played back byte-for-byte from its committed rendition, or
bootstrapped from the agents template when none is committed. Sync never
recomposes it: changing that canon goes through the content skills;CLAUDE.md, the rules under paths.claude_rules,
.claude/settings.json (gzkit hook phases merged into operator-added ones)
and the Claude hook scripts;.github/discovery-index.json and the nested AGENTS.md files;paths.claude_skills and paths.codex_skills, and persona
mirrors under each enabled Claude and Codex surface_root. vendor_skill_map
in src/gzkit/sync_skills.py is the authority for which vendors get a skill
mirror; OpenCode has none.A real run first runs the canonical preflight (refuse_on_sync_blockers,
src/gzkit/commands/sync_guard.py) and exits 1 before writing anything when a
canonical skill is corrupt: a missing SKILL.md or frontmatter, a bad
identity field, a malformed date, version or lifecycle value. A stale
last_reviewed never blocks (GHI #1099). After writing, it appends an
agent_sync_completed ledger event, then re-runs the skill audit and exits 1
if any blocking audit issue other than a stale review remains; the mirrors
are already written by then. Mirror copying only adds and updates files:
mirror-only paths are listed under "Recovery required" and left in place.
Every run prints an Updated <path> line for each surface that always
reports, such as the manifest, AGENTS.md, CLAUDE.md, settings, rules,
hooks and personas, whether its bytes changed or not. Skill mirror files are
listed only when they change.
--dry-run renders the same sync into a capture sink (plan_sync_all) and
writes nothing, the ledger included. It lists the same paths a real run would
report, not only the changed ones. It skips the preflight, so a plan it prints
can still be refused by the real run.
.gzkit/skills/, .gzkit/rules/, .gzkit/personas/,
.gzkit/templates/, .gzkit/chores/, .gzkit/agents/. Never edit a mirror
or src/gzkit/<surface>/.uv run gz agent sync control-surfaces --dry-run, then run it
without the flag.uv run gz skill audit --json and re-run the sync.uv run gz skill audit --jsonuv run gz agent sync control-surfacesuv run gz skill audituv run gz validate --surfaces reports no drift after the sync.uv run gz skill audit shows no mirror-parity error.When Codex is enabled, sync recovers native .codex/hooks.json registrations in
projects carrying scripts/session_orientation.py, and renders registered
.gzkit/agents/roles.json role bodies into .codex/agents/*.toml. Edit the
canonical role body or src/gzkit/hooks/codex.py producer, then preview and sync.
Preserve operator-owned hooks and native role metadata. The generated hook status
labels identify owned handlers; customized unmarked role bodies remain user-owned.
Verify uv run gz validate --surfaces --orientation-freshness. This proves
delivery coherence, not native dispatch: Codex requires review of each new or
changed hook through /hooks before it runs. Report registration, trust, and
observed execution separately. The full parity design remains
ADR-pool.vendor-alignment-codex; syncing does not initiate its parked OBPIs.
These thoughts mean STOP — you are about to create canon/derived drift:
| Thought | Reality |
|---|---|
"I edited .claude/skills/foo/SKILL.md directly — it's the runtime path anyway" | .claude/ is a vendor mirror. The next sync overwrites it. Edit .gzkit/skills/foo/SKILL.md (canonical) and run sync. The instinct to edit the mirror is the bug. |
| "The mirrors look identical to canon — skip sync" | Only a sync regenerates the surfaces from canon, and running it on a synced tree changes no surface bytes. Skipping it after a skill edit leaves the drift in place. |
"Sync will see my edit through the skill-version bump" | Sync copies canon bytes over every mirror and never compares versions; it only checks that the value is X.Y.Z. Bump the version anyway: gz-skill-review requires it, and it is how a reader tells the edit happened. |
| "Sync reported stale mirror-only paths — I'll deal with them next time" | Stale mirror paths mean the canonical source was deleted, renamed or retired. Leaving them means the old skill keeps loading. Run the recovery sequence now. |
"I can skip --dry-run, this edit was small" | The dry-run lists every path the sync would report, before anything is written. It lists unchanged always-reported surfaces too, so read it for the unexpected paths. |
| "Manual copy from canonical to mirror is faster than sync" | Sync also writes the manifest, settings, hooks, nested AGENTS.md files and vendor-specific renderings. A manual copy gets the file and leaves the rest stale. |
.claude/, .agents/ (paths.claude_skills,
paths.codex_skills and each vendor's surface_root in .gzkit.json), or
src/gzkit/<surface>/ in gzkit's own repositoryskill-version bumpuv run gz agent sync control-surfaces --dry-run
uv run gz agent sync control-surfaces
uv run gz validate --surfaces
まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
Use to mine ../airlineops for governance habits gzkit lacks; produces a parity scan report (gzkit leads, AirlineOps adopts).
日本語の概要は準備中です。原文の説明を表示しています。
Use to file a GitHub issue when a finding needs its own work order or the operator asks; OBPI corrections stay in its change log.
日本語の概要は準備中です。原文の説明を表示しています。
Use to do the work a GHI describes, verify it and close the issue with evidence in one sweep.
日本語の概要は準備中です。原文の説明を表示しています。
Use to triage the open GHI queue before planning or choosing what to pull; produces a ranked deliverable.
日本語の概要は準備中です。原文の説明を表示しています。
Run the guarded repository sync ritual; commit-time hooks enforce the cheap gates and the pre-push hook runs gz check.
日本語の概要は準備中です。原文の説明を表示しています。
Run the guarded repository sync ritual; the hooks your pre-commit config declares enforce the quality gates.
日本語の概要は準備中です。原文の説明を表示しています。