sdlc-techdoc

Invoke /techdoc with the feature ID after the SRS exists at the configured srs_dir path. The skill reads the SRS, feature files, and stack profile, then writes TD-{FEAT-ID}.md covering components, data model, interfaces, sequences, and errors.

It requires an SRS markdown file plus its .feature files under the feature directory, and reads the stack profile's layers, layout, and coding standards. A hand-written SRS is valid; missing or unmeasurable NFRs cause a NOT_READY stop.

Yes, when ai_class is not none it writes a prompt spec per prompt name and an eval-set spec with numeric metric thresholds, judge rubrics, and baselines. It also records model, fallback provider, token budget, and guardrails.

ADRs are drafted at Status Proposed for hard-to-reverse triggers such as new providers, major prompt version bumps, RAG adoption, cost budgets, or sending user data to third parties. They are never marked Accepted by the skill.

It does not write implementation code, add new requirements, or weaken NFRs. Missing requirements are routed back to the SRS phase, and unmeetable NFRs are escalated as product decisions rather than designed around.