SDD phase 6 (optional gate) — cross-check constitution, spec, plan, and tasks for consistency and coverage before implementation; writes specs/<id>/analysis.md. Triggers: sdd analyze, consistency check sdd, validate spec plan tasks, sdd gate, pre-implementation review.
SDD — Analyze
TL;DR — Gate on completed tasks (0) → cross-check the four artifacts against each other and the constitution (1) → write analysis.md, mark done or block, route (2). Read <skills-root>/m-sdd/_shared.md first.
Pipeline: tasks → analyze → implement. A skippable gate, not optional decoration: m-sdd-implement refuses to start until analyze is done or explicitly skipped (sdd_set_phase … analyze skipped).
Review discipline for the cross-checks below (this phase does not delegate): verify every claim against the actual artifact files, never a summary or your memory of them; report each finding with artifact + location (section/line) + severity; classify each finding blocking or non-blocking with a one-line rationale.
Phase 1 — Cross-artifact analysis
sdd_set_phase "$root" "$id" analyze in_progress
Compare, and record findings in specs/$id/analysis.md:
Coverage: every spec requirement maps to ≥1 task; flag uncovered requirements.
Test coverage: every user story and success criterion (SC-###) has ≥1
verifying test task ([US#]/[SC-###]). Severity follows the constitution's
testing principle — blocking if it mandates coverage; otherwise a
non-blocking note. If the constitution defines no testing principle, default to non-blocking (coverage expected, not gate-blocking).
Design coverage: if specs/$id/design/ exists, every screen in
design/README.md maps to a user story, and the plan covers every screen. Flag
orphans both ways (non-blocking).
Consistency: plan/tasks don't contradict the spec; names/types used consistently.
Constitution: no artifact violates .sdd/constitution.md (guardrails, conventions).
Scope: still a single implementable unit (else recommend decomposition).
Classify each finding blocking or non-blocking.
Optional companions: before recording the verdict, run m-security-and-hardening to check for security gaps and m-performance-optimization to check for performance regressions. Findings from either feed into this phase's analysis.md as non-blocking notes unless the constitution mandates otherwise.
Phase 2 — Record & route
If any blocking findings: leave analyzein_progress, report them, and send the user back to the relevant phase (m-sdd-specify/m-sdd-plan/m-sdd-tasks).
If clean (or user accepts non-blocking): sdd_set_phase "$root" "$id" analyze done; route to m-sdd-implement. To skip entirely: sdd_set_phase "$root" "$id" analyze skipped.
