gentle-sdd-verify

description: Validate implementation matches specs, design, and tasks

If the native sdd-verify sub-agent is available, delegate this command to it. Otherwise, read the skill file at ~/.claude/skills/sdd-verify/SKILL.md FIRST, then follow its instructions exactly inline.

CONTEXT:

• Working directory: Detect agent-side before proceeding by running git rev-parse --show-toplevel with the Bash tool; if that fails, run pwd with the Bash tool.
• Current project: Derive agent-side from the detected working directory basename. Do not use slash-command shell interpolation for this value.
• Artifact store mode: engram

TASK: Verify the active SDD change. Read the available proposal, specs, design, and tasks artifacts. Then:

This is optional practical verification, including partial diagnostics, with no review-state prerequisite. Report findings without an automatic review/refuter/fix loop. After verify returns, rerun native SDD status and route only from its refreshed nextRecommended. Completed implementation normally proceeds toward archive regardless of diagnostic findings; unfinished implementation normally returns to apply. Do not offer or invoke RDD; do not call gentle-ai review status.

STATUS GATE: Read ~/.claude/skills/_shared/sdd-status-contract.md and produce structured status before acting. If $ARGUMENTS is missing or ambiguous, ask the user to choose and STOP. Do not guess. Missing inputs limit conclusions, not useful partial diagnostics; report missing artifacts and unavailable checks honestly, and never invent passing checks or completed tasks. If status reports workspace-planning, STOP and explain that full workspace implementation verification is not supported in this slice. Carry contextFiles, task progress, dependency states, and actionContext into the native sub-agent prompt when delegating.

ENGRAM PERSISTENCE (artifact store mode: engram): CRITICAL: mem_search returns 300-char PREVIEWS, not full content. Call mem_get_observation(id) for every found artifact; do not invent IDs for missing artifacts. STEP A — SEARCH (get IDs only): mem_search(query: "sdd/{change-name}/spec", project: "{project}") → save spec_id mem_search(query: "sdd/{change-name}/design", project: "{project}") → save design_id mem_search(query: "sdd/{change-name}/tasks", project: "{project}") → save tasks_id STEP B — RETRIEVE FULL CONTENT FOR FOUND ARTIFACTS: if spec_id exists: mem_get_observation(id: spec_id) → full spec if design_id exists: mem_get_observation(id: design_id) → full design if tasks_id exists: mem_get_observation(id: tasks_id) → full tasks Save report: mem_save(title: "sdd/{change-name}/verify-report", topic_key: "sdd/{change-name}/verify-report", type: "architecture", project: "{project}", capture_prompt: false, content: "{verification report}") Set capture_prompt: false when the Engram tool schema supports it; if an older schema rejects or does not expose the field, omit it rather than failing.

Then:

1. Report observed completed and unfinished tasks without changing their state.
2. Check correctness — does code match specs?
3. Check coherence — were design decisions followed?
4. Run tests and build (real execution)
5. Build the spec compliance matrix

Return a structured verification report with: status, executive_summary, detailed_report, artifacts, and next_recommended.