القائمة
Self-audit this repo's skill-first shape - hook suite, word budgets, cross-file duplication, dangling brief paths, version mirrors - and report a scorecard with top fixes
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Fire forge's multi-lens review wave - parallel independent lenses plus a different-model worker over one change. Use when a stop gate says a review wave is owed, or the user asks for crossfire or the review wave by name. Reviews of a branch, a PR, or changes since a commit belong to other review skills. Skip when your own prompt names a run dir and a brief to follow: you are inside a run that fires this wave itself.
Run a code-modifying request end to end - triage, plan, challenge, implement test-first, review wave, fix. Use for any request that changes project code or config - features, fixes, refactors, even small or mechanical ones - instead of editing inline. Skip when your own prompt names a run dir and a brief to follow: you are inside a forge run, and the pipeline never re-enters itself.
Install or update forge on Codex CLI - installs the plugin through the native marketplace, writes forge's enforcement hooks into codex's own hooks file, generates the four tier agents, then live-verifies every declared capability. Run when the startup nag says the installed forge surfaces are stale, or to update forge.
Re-run forge's opencode install/update - refreshes the installed skills, plugin, and tier agents, then re-verifies every declared capability. Run when the startup nag says the installed forge surfaces disagree, or to update forge.
Install forge's bare skill names into ~/.claude/skills - symlink, version-stamped copy where symlinks fail - and write this repo's tracker conventions if absent. Run once after plugin install; re-run when the stale-skills nag says so.
Reflect on the current session to surface workflow friction worth tuning, then audit and curate memory
| name | audit |
| description | Self-audit this repo's skill-first shape - hook suite, word budgets, cross-file duplication, dangling brief paths, version mirrors - and report a scorecard with top fixes |
| disable-model-invocation | true |
Run the deterministic checks below and report a scorecard. Every check is a command whose output decides — no judgment calls, no prose scoring.
python3 -m pytest adapters/claude-code/hooks/tests -q from the repo root. Green = pass; a failure names its own fix.skills/**/*.md, adapters/*/skills/**/*.md, and .claude/skills/**/*.md fits one read: ~865-word target, 2,000-word hard ceiling. wc -w each file; over ceiling = fail, over target = warn. List offenders with their counts.NAME.md / ../dir/NAME.md sibling path exists on disk, resolved relative to the referencing file. A missing target = fail.python3 scripts/bump-version.py --check (also pinned by the pytest gate). Report the line.