一键导入
这个仓库中的 skills
Mandatory onboarding. Every agent MUST execute this on startup, or as soon as becoming aware of it.
Execute an active implementation phase against DR/IP intent, keep notes current, and surface blockers early. You MUST use this skill when implementing a delta / IP.
Before starting anything new, understand task, intent, and context.
Understand project governance to avoid spreading heresy. you
Scope intentional change as a delta. Define applies-to specs/requirements, risks, and closure targets before implementation.
Shape a spec revision when requirements/responsibilities move. Use this before delta scoping when policy/doctrine requires revision-first flow.
Perform verification/audit pass after implementation (or as discovery/backfill), reconcile against specs/contracts, and route findings into closure or follow-up change.
Invoke this skill whenever any of the following occurs during work: (1) you discover or confirm a durable fact, constraint, invariant, or “how we do X here” pattern; (2) you create a new workflow/checklist that will be reused; (3) you resolve a recurring confusion (“where is the source of truth?”); (4) you make a decision that is not an ADR/SPEC but would prevent rework; (5) you detect a sharp edge, footgun, or non-obvious dependency that future agents will hit. Do NOT rely on conversational context to persist. When the information would save ≥10 minutes for a future agent, write a memory record immediately.
Close a delta safely - satisfy coverage gates, complete the delta command, and verify owning-record lifecycle updates.
Identify and address obstacles, significant decisions, or emergent complexity. You MUST use this skill if you encounter unanticipated obstacles during implementation.
Write a prompt to help the next agent continue effectively.
Draft or refine a design revision (DR) for a delta, capturing current vs target behavior, code impacts, and verification alignment.
implement a well-defined task or implementation plan
Invoke this skill whenever you observe memory drift or when your actions would invalidate existing memories. Mandatory triggers: (1) you change a workflow/command, move files, rename modules, or change invariants; (2) you discover a memory is wrong, missing provenance, or stale; (3) you see a memory record guiding behaviour that is no longer true; (4) you find duplicates or near-duplicates; (5) you are about to add a new memory that overlaps an existing one. Core rule: if you change reality, you must change memory in the same change-set (or immediately after) so future agents do not inherit incorrect guidance.
Whenever you complete a task or phase - record implementation notes.
Plan execution for a delta - refine IP objectives/gates and create the next phase sheet with concrete tasks and verification expectations.
Invoke this skill before making non-trivial assumptions in a large codebase. Mandatory triggers: (1) you are about to modify a subsystem you have not touched in this run; (2) you are about to run, change, or suggest a command pipeline (tests, builds, releases, migrations); (3) you see conflicting cues in code/docs; (4) you are asked “what is the right way here?”; (5) you are debugging a recurring failure mode; (6) you are about to answer with “probably/usually/likely”. Default rule: if you cannot cite a source-of-truth file/doc/ADR/SPEC from the repo, you must consult memories first and then proceed.
Invoke this skill as a deliberate review pass whenever stability matters: (1) before a release/migration/large refactor; (2) at the start of work in an unfamiliar subsystem; (3) when you see repeated agent confusion; (4) when thread-type working sets accumulate; (5) when you suspect stale guidance is causing defects. This is not ad-hoc maintenance; it is a structured audit to prevent systemic drift.