菜单
Deep audit of skill quality against current skill-creator best practices. Flags missing when_to_use, imperative density, missing evals/, and description budget overflow.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Build Flutter APK/IPA, generate changelog, update buildlog, and commit version bump. Handles platform selection, build mode, and artifact renaming.
Build Unity APK/IPA on macOS with Unity 6.x — pre-flight checks, version bump, iOS signing params, post-build verification, buildlog, and commit.
Evidence-driven bug investigation and fix pipeline. Use when something is broken, not working, or the root cause is ambiguous.
Two independent reviewers must both approve high-stakes output before it ships, breaking the single-reviewer blind-spot problem.
Team-orchestrated feature development with parallel specialist agents. Use when building a feature that benefits from multiple agents working concurrently. You act as Product Director coordinating the team.
Interactive agent picker for composing and dispatching parallel teams. Browse available agents across domains, pick a roster, dispatch in parallel.
| name | skill-audit |
| description | Deep audit of skill quality against current skill-creator best practices. Flags missing when_to_use, imperative density, missing evals/, and description budget overflow. |
| when_to_use | Use after writing or modifying a skill, before a release, or when reviewing skill quality. Triggers: "audit my skills", "check skill quality", "are my skills well-written", "/skill-audit", "skill best practice check", or after adding new SKILL.md files. |
| user-invocable | true |
| argument-hint | [plugin-name] |
| allowed-tools | Bash Read Glob Grep |
Runs validate.js --audit to surface skill-quality issues that the standard /validate skips. Specifically:
when_to_use and inline trigger phrases (combat undertriggering per skill-creator's "pushy descriptions" guidance)MUST/ALWAYS/NEVER occurrences (reframe stylistic imperatives as prose)feature-dev, bug-fix, cleanup, clarify, team-feature-dev, product-explore) without evals/evals.json (needed for skill-creator's run_loop.py description optimization)description + when_to_use exceeds the canonical 1,536-character capPlus everything /validate already checks (frontmatter validity, body length, model+effort assignment, antipatterns).
# Full audit, all plugins
node "${CLAUDE_PLUGIN_ROOT}/scripts/validate.js" --root . --check skills --audit
# Single plugin
node "${CLAUDE_PLUGIN_ROOT}/scripts/validate.js" --root . --check skills --audit --plugin nextc-core
# JSON output for tooling
node "${CLAUDE_PLUGIN_ROOT}/scripts/validate.js" --root . --check skills --audit --json
The PreCommit hook (configured in ~/.claude/settings.json) blocks commits with WARN/FAIL findings. You can run /skill-audit proactively whenever you:
Each finding has:
nextc-core)feature-dev)Treat any FAIL as blocking. WARNs are advisory but in aggregate signal description budget waste, undertriggering risk, or skill-creator best-practice drift.
~/.claude/plugins/marketplaces/claude-plugins-official/plugins/skill-creator/skills/skill-creator/SKILL.mdnextc-claude-toolbox/scripts/validate.js (audit-mode block at S09–S12)