ワンクリックで
shared-agents
shared-agents には mrlunchbox777 から収集した 13 個の skills があり、リポジトリ単位の職業カバレッジとサイト内 skill 詳細ページを表示します。
このリポジトリの skills
Use for complex software tasks that need interactive main-session planning followed by structured checklist execution.
Use to validate agent skill structure, frontmatter, path/name consistency, required sections, and broken references before skill changes are accepted.
Use to audit the skill ecosystem for capability coverage, overlapping responsibilities, fragmented guidance, and consolidation opportunities.
Use to evaluate agent skills with should-trigger and should-not-trigger cases, executable validation hooks, and pass/fail reports for review or CI.
Use when the user asks to create, improve, copy, port, evaluate, or brainstorm an agent skill. Probes the desired workflow, researches local and upstream skill examples, recommends patterns, then invokes plan-and-execute for implementation.
Trigger keywords: "grill me", "grill-me". Use only when the user explicitly requests grill-me style relentless, one-question-at-a-time interviewing.
Use to pressure-test implementation plans against codebase docs, refine language, capture ADR decisions, and emit machine-readable handoff JSON for downstream workflows.
Use to draft ADR records for strict-threshold decisions discovered during grilling and produce blocker-ready issue metadata.
Use to run risk-triggered architecture follow-up reviews and maintain ARCHITECTURE_REVIEW.md cadence metadata.
Use to convert a ready grilling outcome into machine-linked implementation issues with required schema and blocker dependencies.
Use to refine and persist shared domain terms in CONTEXT.md (or selected context docs), emit terminology deltas, and enforce deprecated-term policy.
Use ONLY when the user asks to initialize or bootstrap opencode global setup (for example "init", "first-time setup", "configure ~/.config/opencode", or "set up shared-agents"). Verifies shared-agents root, proposes global AGENTS.md and opencode.json updates, and requires explicit confirmation before writes.
Use when user asks for <trigger keywords, file names, or workflow>. Include when to use and when not to use.