Instructions file CodexOpenCode
AGENTS.md instructions for skyfiredao/aia_kaiwu, covering agents.md, 1. route every turn, 2. intent gate, 3. investigate before asserting and 4. pragmatism and scope.
Instructions file CodexOpenCode
AGENTS.md instructions for skyfiredao/aia_kaiwu, covering agents.md, 1. route every turn, 2. intent gate, 3. investigate before asserting and 4. pragmatism and scope.
Instructions file
Claude Code instructions for skyfiredao/aia_kaiwu, a project described as: A workflow-oriented agent system that runs the same set of engineering roles across modern AI coding assistants (opencode, Cursor, Claude Code, VS Code + Copilot).
Skill Claude CodeCodex
Align documentation with the actual code after iterative changes. Use when docs may be stale, missing, or drifted from implementation.
Skill Claude CodeCodex
Execute an approved plan end-to-end. Use when a plan already exists and needs implementing (single-scope or multi-module).
Skill Claude CodeCodex
Hard-bug diagnosis end-to-end. Use for flaky, fix-resistant, or chained-symptom bugs with no obvious cause.
Skill Claude CodeCodex
Bug diagnosis, first pass. Use for a reported bug with a clear reproduction; escalates to job-debug-deep if it resists a straightforward fix.
Skill Claude CodeCodex
Turn an ambiguous or multi-step request into an approved, written plan. Use before implementing when scope is unclear or spans multiple steps.
Skill Claude CodeCodex
Maintain an open pull request through its review lifecycle. Use to address reviewer comments, update the PR description, and reply to review threads.
Skill Claude CodeCodex
Comprehensive pre-completion review. Use before claiming work done, to check the diff against the plan and standards.
Skill Claude CodeCodex
Build a comprehensive solution through systematic multi-angle (SCAMPER) analysis, round by round until no new angle remains. Domain-general, aimed at a comprehensive solution, not fault-finding. Triggers - "scamper", "multi-angle analysis", "build a comprehensive solution from all angles", "全角度分析".
Skill Claude CodeCodex
End-to-end security audit. Use for security review or when work touches auth, crypto, or secrets.
Skill Claude CodeCodex
Single-session autonomous supervisor that drives plan to execution to review to Done with no interaction, inside a git worktree. Use for "spiral up", "run to convergence", "loop until done", "循环直到结束".
Skill Claude CodeCodex
Verify in-progress work against the approved plan. Use mid-execution to check progress.
Skill Claude CodeCodex
Adversarial review of a written plan or design document for clarity, completeness, and verifiability. Use when one already exists and needs a rigorous pass before implementation.
Skill Claude CodeCodex
Mid-execution checkpoint review. Use after every 3 steps, before major state transitions (schema migrations, deployments, git push), or when the executor reports drift.
Skill Claude CodeCodex
Critique the user's request itself, before planning. Use when the request is ambiguous, sounds larger than one turn, or contains contradictions. Never proposes a plan; only critiques the request.
Skill Claude CodeCodex
Hard-bug diagnostics. Use for hard bugs, perf regressions, "sometimes fails" flakiness, or before a second fix attempt the executor cannot pin down.
Skill Claude CodeCodex
Distill an overview-level project spec (glossary, domain model, design principles, business logic) from existing docs into a preloadable AGENTS.md-style file. Compress and define only; never infer new design. Triggers — "蒸馏项目 spec", "生成项目 overview/术语基线", "distill a project overview for the team to preload".
Skill Claude CodeCodex
Write documentation from code and context. Use for API reference, architecture doc, guide, or README.
Skill Claude CodeCodex
Contextual grep inside this repository. Use to answer "where is X defined?", "which file handles Y?", or to find a cross-cutting pattern.
Skill Claude CodeCodex
Coordinator for multi-step work that mixes research and execution across several unrelated systems or domains.
Skill Claude CodeCodex
High-authority audit on two independent axes (Standards + Spec). Use for architectural verdicts, hard debugging after ≥2 failed fix attempts, post-implementation review before claiming done, or security/performance judgment calls.
Skill Claude CodeCodex
External references, docs, OSS repositories, official APIs, best-practice research. Primary sources only. Use for "how do I use library X?", "what's the best practice for Y?", "how does OSS repo Z do this?".
Skill Claude CodeCodex
Multi-domain implementer that delegates to specialists. Use when a task spans multiple modules. For single-scope tasks, prefer task-yugong.