Analyze

User Input

$ARGUMENTS

Outline

$ARGUMENTS (if any) narrows scope: artifact (spec/plan/tasks), category (coverage/ambiguity/azure), or a T###. Empty = analyze everything.

Strictly read-only. Never modify files; only produce a report.

1. Load context. Read specs/feature.json → <feature_directory>. Read constitution.md, spec.md, plan.md, tasks.md, tasks.json. Stop if any is missing and name the command to run first (/speckit:specify / /speckit:plan / /speckit:tasks). If tasks.md and tasks.json disagree (IDs, descriptions, order), record as CRITICAL and stop.

2. Build a lightweight model (don't echo raw artifact text in the report):

   • Functional requirements + success criteria from spec.md, keyed by explicit IDs (FR-001, SC-002) when present, else imperative slug.
   • Architecture decisions, Azure services, non-negotiables from plan.md + constitution.
   • Tasks from tasks.json with depends_on, parallel_with, satisfies, files, status.

3. Detection passes — high-signal only, cap 50 findings, summarize overflow:

   • Constitution alignment — any element conflicting with a non-negotiable. Always CRITICAL.
   • Coverage — requirements/criteria with no task in satisfies; tasks whose satisfies references nothing in spec.
   • Duplication — near-duplicate requirements or tasks.
   • Ambiguity — vague terms (fast, secure, scalable, robust) without measurable criteria; unresolved markers (TODO, ???, [NEEDS CLARIFICATION: …], Resolve: …).
   • Underspecification — tasks without files; requirements with verbs but no measurable outcome; user stories missing acceptance criteria.
   • Inconsistency — terminology drift; entities in plan but not spec (or vice versa); dependency cycles or out-of-order depends_on.
   • Azure readiness — services in plan.md with no provisioning task; identity/RBAC, networking, region, or data-residency constraints unaddressed; secrets/connection strings inline instead of via Key Vault / managed identity; missing IaC entry point (azure.yaml, infra/main.bicep, main.bicep, infra/main.tf, main.tf) when plan declares Azure resources.

4. Severity:

   • CRITICAL — violates a non-negotiable, blocks baseline functionality, or breaks tasks.md/tasks.json sync.
   • HIGH — conflicting requirements, ambiguous security/perf attribute, untestable acceptance criterion, missing provisioning for a declared Azure service.
   • MEDIUM — terminology drift, missing non-functional coverage, underspec'd edge case, missing IaC for declared resources.
   • LOW — wording, minor redundancy.

5. Emit the report as Markdown only (no file writes):

   ## Specification Analysis Report
   
   | ID | Category | Severity | Location(s) | Summary | Recommendation |
   |----|----------|----------|-------------|---------|----------------|
   | C1 | Constitution | CRITICAL | plan.md §Architecture | … | … |
   
   ### Coverage
   | Requirement / Criterion | Has Task? | Task IDs | Notes |
   |-------------------------|-----------|----------|-------|
   
   ### Unmapped Tasks
   - T### — description (no `satisfies`)
   
   ### Azure Readiness
   - Service / constraint → status (covered / missing IaC / missing task)
   
   ### Metrics
   - Requirements: N
   - Tasks: N (pending: N, done: N)
   - Coverage: N% (requirements with ≥1 task)
   - Ambiguity findings: N
   - Duplication findings: N
   - CRITICAL findings: N
   

   Use stable finding IDs by category initial: C constitution, V coverage, D duplication, A ambiguity, U underspec, I inconsistency, Z azure.

6. Next actions. If any CRITICAL, recommend resolving before /speckit:implement or /speckit:deploy and name the command(s) to run. If only LOW/MEDIUM, note user may proceed and list suggestions. Ask whether to produce concrete remediation suggestions for the top findings — never apply edits.

Guidelines

• Read-only — never modify files.
• Don't invent missing sections; report them as findings.
• Constitution non-negotiables outrank everything else — only addressable by changing spec/plan/tasks or via /speckit:constitution.
• Cite specific locations (file + section or task ID), not generic patterns.
• Rerunning on unchanged artifacts must produce identical finding IDs + counts.