Help us improve
Share bugs, ideas, or general feedback.
From harness
Finalizes verified features by merging spec/design deltas into full docs, running code review, unifying style, splitting atomic commits, pushing, and triggering self-evolution. Use after /verify PASS.
npx claudepluginhub zxdxjtu/harness --plugin harnessHow this skill is triggered — by the user, by Claude, or both
Slash command
/harness:archiveThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
You are finalizing a completed feature: merging incremental specs back into the full documents, running a final code review, making clean commits, and triggering self-evolution.
Orchestrates end-to-end feature workflow from requirements to design, tasks, build, verify, and archive using sub-agents in one command.
Orchestrates multi-session projects by implementing one feature per cycle from feature-list.json through TDD pipeline with quality gates and code review.
Share bugs, ideas, or general feedback.
You are finalizing a completed feature: merging incremental specs back into the full documents, running a final code review, making clean commits, and triggering self-evolution.
Parameter: $ARGUMENTS — feature ID (e.g. F001)
cat .harness/evidence/${FEATURE_ID}/verdict.md 2>/dev/null
The feature must be verified (verdict = PASS) before archiving. If not verified → suggest running /verify first.
Read .harness/specs/{FXXX}-*.md (the incremental spec).
Read .harness/full-spec.md (the full project spec, if exists).
Merge the feature spec into the full spec:
If conflicts detected (e.g., feature contradicts existing spec), present to user for resolution.
Read .harness/designs/{FXXX}-*.md (if exists).
Read .harness/full-design.md (if exists).
Merge similarly: add new components, update dependency graph, add new data models.
Launch a comprehensive code review sub-agent:
Agent(subagent_type: "code-reviewer", run_in_background: true,
prompt: "Final code review for feature {FXXX}.
Review ALL files changed for this feature:
git diff {base_commit}..HEAD -- {files from tasks.md}
Check:
1. Code quality: naming, readability, complexity
2. Error handling: no swallowed errors, proper messages
3. Security: no hardcoded secrets, input validation
4. Immutability: no mutation of shared state
5. Style consistency: matches project conventions
Report: CRITICAL / HIGH / MEDIUM issues only.
Ignore cosmetic issues.")
Read the review. If CRITICAL issues → fix before proceeding. HIGH issues → fix or acknowledge.
If a lint/format command is configured:
{lint_command} --fix 2>&1
{format_command} 2>&1
Commit style fixes separately: chore(style): unify code style for {FXXX}
Review the git history for this feature. If commits are messy or mixed:
Present the commit plan to user:
"当前有 {N} 个 commits。建议整理为 {M} 个原子 commits: {list}。是否同意?"
If user agrees, execute interactive rebase (only if user explicitly approves).
git push origin {current_branch}
Ask user:
"代码已推送。是否需要创建 MR/PR?"
If yes:
gh pr create --title "feat({FXXX}): {feature name}" --body "$(cat <<'EOF'
## Summary
{feature description}
## Changes
{list of key changes}
## Test Evidence
- V1 (Unit): ✅ {pass count}/{total}
- V2 (Integration): ✅ {pass count}/{total}
- V3 (E2E): ✅ {pass count}/{total}
- Evaluation Score: {score}/10
## Evidence
See `.harness/evidence/{FXXX}/` for full reports.
EOF
)"
Record this iteration for self-evolution analysis:
Append to .harness/evolution/memory.jsonl:
{"iteration": N, "feature": "FXXX", "name": "{feature name}", "date": "{ISO}", "complexity": "{complexity}", "waves": M, "tasks": K, "eval_score": X.X, "lessons": [], "process_improvements": []}
Update config iteration count:
"evolution": { "iteration_count": N+1 }
If this is the 2nd or later feature completed, ask:
"已完成 {N} 个 feature 迭代。是否要运行自进化分析(检查模板和流程是否需要优化)?"
If user agrees → execute /evolve logic.
📋 SDD 进度 — {FXXX}: {feature name}
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
✅ proposal → ✅ tdd-align → ✅ decompose → ✅ sprint → ✅ evaluate → ✅ verify → ✅ archive
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🎉 Feature {FXXX}: {feature name} 已完成全部 SDD 流程!
评估得分: {score}/10 | 任务: {completed}/{total} | Waves: {M}
"Feature 已归档完成。是否要开始下一个 Feature?"