メニュー
Checks klank's agent setup for consistency - mirror parity, frontmatter, cross-refs, stale paths, descriptions. Use before any commit under .claude/, .github/agents/, or docs/agents/.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Creates a new self-contained subagent identity in .claude/agents/ and its .github/agents/ (Copilot) and .junie/agents/ (Junie) mirrors. Use when introducing a new expert identity to the klank agent system.
Diagnoses a failing CI/CD run via failed-step logs, debug logging, re-runs, flaky triage, and local repro. Loads a platform reference (e.g. github-actions). Use when a pipeline run is failing.
Audits and hardens an existing CI/CD pipeline against supply-chain and injection attacks. Loads a platform reference (e.g. github-actions) with a full checklist. Use when securing pipelines.
Keeps CI/CD pipelines healthy via dependency and runner-image updates, pruning, and reusable-pipeline versioning. Loads a platform reference (e.g. github-actions). Use for routine upkeep.
Scaffolds a secure-by-default CI/CD pipeline - least-privilege, pinned deps, caching, matrices, reusable parts. Loads a platform reference (e.g. github-actions). Use when creating a pipeline.
Makes a product design inclusive at the design stage - WCAG 2.2 POUR, inclusive design across impairments, an a11y checklist. Use when designing for or auditing accessibility.
| name | audit-agent-setup |
| description | Checks klank's agent setup for consistency - mirror parity, frontmatter, cross-refs, stale paths, descriptions. Use before any commit under .claude/, .github/agents/, or docs/agents/. |
.claude/, .github/agents/, or docs/agents/add-role or add-skillRun each check in order; fix all failures before committing.
.claude/agents/<n>.md has a matching .github/agents/<n>.agent.md and .junie/agents/<n>.md, and vice versa. Skills are not mirrored: flag any .github/agents/<n>.agent.md that shadows a .claude/skills/<n>/ directory (Copilot auto-discovers skills).SKILL.md has name (kebab-case, ≤ 64 chars) and description (≤ 1024 chars, leading verb). Every .claude/agents/<n>.md has name, description, model, and either tools or disallowedTools.## Skills used bullet resolves to a real .claude/skills/<n>/ directory..claude/agents/<n>.md and .github/agents/<n>.agent.md body is a real identity, not a redirect (Read … and follow) to a parallel doc. There is no role-routing table in CLAUDE.md/AGENTS.md.description are identical across .claude/agents/<n>.md, .github/agents/<n>.agent.md, and .junie/agents/<n>.md; model matches between the .claude/ and .github/ copies (the .junie/ copy carries only name + description).agent-setup.md §Skill Catalogue exists as .claude/skills/<n>/SKILL.md on disk; and every .claude/skills/ directory has a catalogue row..claude/, .github/agents/, or docs/agents/ markdown resolves on disk.SKILL.md description leads with a third-person verb, says what + when, and its first sentence is ≤ 200 chars.See references/checks.md for verbose explanations and fix-up examples.
add-role. Check 1 (stray skill mirror) → delete the .github/agents/ stub; the skill stays in .claude/skills/.< or > are patterns, not real paths - skip them.