rqmd-init

Files (1)

• SKILL.md

SKILL.md (first lines)

Onboard a repository into rqmd through a structured multi-choice interview with smart defaults. For existing repos, explore the codebase first — the interview surfaces what actually matters in this project — then write the scaffold from confirmed answers.

## Done when

- `docs/requirements/` exists with an index and at least one seeded domain file
- `/dev` and `/test` skills populated with the repo's confirmed build/run/test commands
- ID prefix confirmed and recorded in the requirements index
- `rqmd --verify-summaries --non-interactive` passes

## Interview flow

Conduct as a structured multi-choice conversation with smart defaults pre-selected:

1. **Repo context** — language, framework, build system; run `gh issue list --limit 20` when `gh` is available
2. **ID prefix** — suggest a short project-specific prefix (3-5 chars); avoid generic fallbacks like `REQ` or `RQMD`
3. **Requirement domains** — propose up to `max_domain_files` domain files from discovered source areas; let user confirm/remove/add
4. **Dev skill** — discover build and run commands; **read `.vscode/tasks.json` and `.vscode/launch.json` if present** and surface their labels in the generated `skills/dev/SKILL.md` so the agent prefers them over raw shell
5. **Test skill** — discover test commands and smoke path; **note whether the project has a Test Explorer-discoverable framework (pytest, vitest, jest, mocha, go test, …)** and instruct the agent in `skills/test/SKILL.md` to prefer the `runTests` tool over raw shell invocations
6. **Seed requirements** — draft P2 stubs for discovered features and open issues (max `max_issue_requirements` from issues; max `max_source_areas` source areas)

Then write: run `rqmd --scaffold` (or command palette "rqmd: Initialize Project") for the requirements index; write `/dev`, `/test` skills and domain seed files directly.

## Edge cases

- Treat all seeded requirements as starting points — mark P2 and adjust priorities after review
- For repos with existing `docs/requirements/`: do not overwrite; offer to add new domain files alongside
- Continue gracefully when `gh` is unavailable or unauthenticated
- Keep first write small: respect `max_domain_files` and `max_issue_requirements` limits

## Telemetry instrumentation

When a named friction point below occurs, call `send_event` **inline** — do not load `/rqmd-telemetry` first:

| Friction point | `step` | Severity |
|---|---|---|
| `rqmd` command not found | `preflight` | `high` |
| `rqmd --scaffold` (or `rqmd: Initialize Project`) fails | `scaffold` | `medium` |
| `rqmd --verify-summaries --non-interactive` fails after scaffold | `post_scaffold_verify` | `low` |

```python
from rqmd.telemetry import send_event
send_event(
    event_type="struggle",
    severity="...",  # from table above
    summary="...",  # one-line description
    detail={
        "workflow": {"skill": "rqmd-init", "step": "..."},
        "command": "...",
        "stderr_snippet": "<first 500 chars>",
    },
)
```

> Payloads are scrubbed for secrets and PII (RQMD-TELEMETRY-016) before transmission. Truncate `stderr_snippet` to 500 chars yourself.

From ryeleo/rqmd-vscode, MIT.

More from ryeleo/rqmd-vscode