ai-engineering-skills

Choose the smallest workflow that still buys back enough attention and safety for the task.

Record clear evidence that a task works, including commands, results, and remaining risks.

Compress project context into HANDOFF.md with workflow state, active modes, next gate, verification, current hypothesis, and resume packet for the next agent session.

Implement exactly one planned slice after scope is frozen.

Detect context drift, rehydration loops, hypothesis sprawl, scope creep, active-mode loss, and durable-state gaps. Recommend the smallest corrective action.

Create a compact SPEC.md that prevents premature agreement by defining intent, non-goals, likely failure modes, and verification evidence.

Explicitly limit blast radius before any implementation or file changes.

Add focused deterministic tests, fixtures, or smoke checks.

Run a lightweight GO or NO-GO gate before use, especially for autonomous or scheduled workflows.

Run a short boundary-grilling pass for small AI-engineering tasks, clarifying intent, non-goals, terms, constraints, and readiness before mini-spec.

Compress routine agent responses to reduce token usage while preserving commands, file paths, assumptions, risks, verification details, and next actions.

Check consistency across `CONTEXT.md`, `SPEC.md`, `PLAN.md`, `TODO.md`, and scope-freeze notes.

Preserve discovered bugs as BUGS.md entries.

Validate a mini-spec before planning.

Create a small set of project rules before repeated agent work.

Debug failures systematically.

Convert a mini-spec into 3-7 vertical implementation slices.