Reusable skills for coding agents

Turn a rough product idea into the first task inventory: user goal, scope, non-goals, risks, unknowns, dependencies, and decision points.

Use it before writing specs when the problem is still blurry.

Convert an existing prototype into a spec-driven sequence: capture behavior, extract requirements, define acceptance criteria, then produce implementation tasks.

Use it when a working demo exists but the team needs maintainable specs before scaling.

Create a project index that links README, architecture notes, OpenSpec changes, current tasks, accepted decisions, and verification commands.

Use it when agents keep losing project context or repeating stale decisions.

Document agent permissions, tool boundaries, model routing, secrets policy, handoff rules, rollback paths, and human approval checkpoints.

Use it before connecting multiple agents, MCP tools, cloud runners, or provider routers.

Write a change proposal before implementation: problem, proposed behavior, affected specs, migration notes, acceptance tests, and review questions.

Use it for features that are too large for a single prompt.

Split a spec into agent-sized tasks with owned files, verification commands, dependencies, and merge order.

Use it when multiple agents or multiple sessions will work on the same feature.

Capture architecture decisions, alternatives rejected, tradeoffs, date, owner, and the specs or tasks affected by the decision.

Use it when agents keep reopening already-settled design questions.

Define what counts as done: tests, build, browser check, migration dry run, screenshots, logs, and reviewer evidence.

Use it before an agent claims a task is complete.

Specify which tasks can use cheaper models, which require frontier models, and how fallback, token compression, and audit logs should work.

Use it when AI coding volume creates cost, quota, or reliability pressure.