| name | rwyn |
| description | Use when working in a repository that uses `rwyn` to build graphs, verify a change, plan steps from diffs, explain why steps were chosen, dry-run the execution plan, or reason about repo-local `.rwyn/config.yaml`. Trigger for requests about change impact analysis, verification, step planning, `rwyn verify`, `rwyn plan`, `rwyn explain`, `rwyn run`, repo-local config, declarative `plugins:` transforms, or dogfooding CI/planning behavior. |
Rwyn
rwyn means "run what you need". Use this skill when the user wants to understand or operate the rwyn workflow in a repo.
Workflow
- Run from the target repo root when possible.
- Check for
.rwyn/config.yaml or .rwyn/config.yml. If present, prefer config-backed commands without repeating --catalog, --graph, or --policy.
- Start with
rwyn doctor if the repo setup is unclear.
- Use the smallest command that answers the question:
- use
rwyn setup and $setup when the repo needs initial config or transform scaffolding
rwyn verify for the beta day-to-day loop: plan, inspect coverage, refresh stale or missing scoped evidence, replan, and optionally run
rwyn plan for what would run without maintaining coverage
rwyn explain <file> for why a file selects certain steps
rwyn run --plan for the layered execution plan
rwyn run --dry-run for the exact commands
rwyn compare and rwyn ingest learn for run-record analysis
- When a result feels wrong, inspect it with
rwyn explain before changing policy or engine code.
Command Guide
- Health check:
- Get the agent setup contract:
- Build or refresh the graph:
- Plan steps for the current branch diff:
rwyn verify --base origin/main --dry-run
rwyn plan --base origin/main
- Plan steps for local edits:
- Maintain coverage evidence and run the final plan:
rwyn verify --base origin/main --run
- Explain one file:
rwyn explain path/to/file.sol
- Show the execution plan:
rwyn run --uncommitted --plan
- Show the concrete commands without running them:
rwyn run --uncommitted --dry-run
See references/commands.md for a fuller command cookbook and interpretation notes.
Interpretation
signal is a planner score, not a bug severity score.
- Coverage is strong evidence for test scoping, but not the whole selection model.
rwyn verify should make coverage maintenance feel automatic. Stale and missing scoped evidence should be refreshed before the final plan is trusted, while uncovered evidence should remain useful evidence rather than causing rerun loops.
- Import reachability is a fallback signal. Treat it as heuristic, not truth.
- Declarative
plugins: transforms are the right place for repo quirks like interface/source links or generated bindings.
- Text-only changes should usually suppress semantic steps except for steps intentionally marked as text-sensitive.
Output Expectations
- Prefer concise summaries over pasting massive JSON unless the user asked for raw output.
- For large scoped items, summarize the count and a short preview of scopes.
- If
rwyn selects something surprising, name the exact step and the exact evidence path that caused it.