菜单
Push the current free branch, fork xp-close-reviewer for free-mode review, and merge into the target integration branch (recorded plan branch when set and present locally, else primary) with cleanup. Used for ad-hoc work performed outside a sprint or plan; degrades gracefully when gh is unavailable.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | xp-free-close |
| description | Push the current free branch, fork xp-close-reviewer for free-mode review, and merge into the target integration branch (recorded plan branch when set and present locally, else primary) with cleanup. Used for ad-hoc work performed outside a sprint or plan; degrades gracefully when gh is unavailable. |
| allowed-tools | ["Read","Write","Bash","Bash(*/append.sh *)","Bash(*/init.sh)","Bash(*/skills/*/scripts/*)","Bash(python3 */scripts/branching.py *)","Bash(python3 */scripts/close_common.py *)","Bash(git push *)","Bash(gh pr *)"] |
!CLAUDE_PLUGIN_DATA="${CLAUDE_PLUGIN_DATA}" ${CLAUDE_SKILL_DIR}/scripts/preload.sh
Sequential discipline. The harness batches independent tool calls in parallel; this skill is step-gated. Run Step 1 → 2 → 3 → 4 → 4b → 4.5 → 5–6 → 7 strictly, one step per turn — make the call, observe, then decide the next. Don't batch a step with the one that depends on it (e.g. pushing or merging before the forked xp-close-reviewer returns); never spawn the same subagent twice. Independent read-only calls may still batch.
The preload above surfaces SMM_DIR, CURRENT_BRANCH, TARGET_BRANCH, GH_AVAILABLE, and WORKTREE_CLEAN. TARGET_BRANCH is the recorded plan branch when execution_plan.json sets one (and that branch exists locally), else the primary integration branch (sprint-close parity) — a free branch forked off a plan branch merges back to the plan branch, never to main. Shared pipeline lives in ${CLAUDE_PLUGIN_ROOT}/scripts/close_common.py.
Commit trailer reminder — this is the ONLY leg that closes a Try. When a commit on this free branch lands a carried retro Try (one adopted via /xp-work-selection adopt), the commit body must include Resolves-Event: <try-id>. Adoption now links its Try rather than closing it — deliberately, so that taking work on cannot close the item that verifies the work landed. There is no cascade to fall back on: without the trailer, the Try stays open forever and the retro agent keeps carrying it as adopted-but-never-landed.
python3 ${CLAUDE_PLUGIN_ROOT}/scripts/close_common.py preflight \
--cwd . --current <CURRENT_BRANCH> --target <TARGET_BRANCH>
If the exit code is non-zero, stop — script emitted the reason on stderr (dirty worktree or <CURRENT_BRANCH> IS <TARGET_BRANCH>).
python3 ${CLAUDE_PLUGIN_ROOT}/scripts/close_common.py push \
--cwd . --branch <CURRENT_BRANCH>
PR_OUTPUT=$(python3 ${CLAUDE_PLUGIN_ROOT}/scripts/close_common.py create-pr \
--cwd . --base <TARGET_BRANCH> --head <CURRENT_BRANCH> \
--title "<short title for the free work>" --body "<one-paragraph summary>")
PR_OUTPUT is a PR number or skipped: no gh on PATH.
Apply the shared ### Step 4: Security Review block above with <close-mode> → free and <close-skill-name> → xp-free-close.
Apply the shared ### Step 4b: Full code review (conditional) block above.
Compute the diff command:
DIFF_CMD=$(python3 ${CLAUDE_PLUGIN_ROOT}/scripts/close_common.py diff-command \
--target <TARGET_BRANCH> --source <CURRENT_BRANCH>)
Substitute the captured value as <DIFF_CMD> in the agent prompt below.
Agent(
subagent_type: "xp-agents:xp-close-reviewer",
prompt: "SMM_DIR=<SMM_DIR>\nSYSTEM_CONTEXT_RENDERED=<SYSTEM_CONTEXT_RENDERED>\n\n## Mode\nfree\n\n## Source Branch\n<CURRENT_BRANCH>\n\n## Target Branch\n<TARGET_BRANCH>\n\n## Diff Command\n<DIFF_CMD>\n\n## Close Cycle ID\n<CLOSE_CYCLE_ID>\n\n## Context\nClosing free branch <CURRENT_BRANCH> into <TARGET_BRANCH>. PR <PR_OUTPUT or 'not created (no gh)'>.\n\n## Instructions\nRun the Diff Command, analyze cumulative diff with free-mode focus (standard quality review only — code quality, clarity, naming, obvious bugs, missing error handling at boundaries, test coverage gaps). Sprint/plan-level scope concerns do not apply. Return Keep / Concern / Block summary."
)
The shared close-pipeline reference (Steps 5, 5b, and 6) is emitted by the preload at the top of this context — see scripts/_close_pipeline_shared.md for the source. Apply those three steps in order after Step 4.5, then continue with Step 7 below.
Free-close override for Step 6 (auto-merge gate): if ALL of these hold, skip the shared Step 6's AskUserQuestion and proceed directly to Step 7:
Step 5c queued zero ask-user items. Verify via the canonical structured filter:
ASK_COUNT=$(python3 ${CLAUDE_PLUGIN_ROOT}/smm/smm_cli.py \
--smm-dir <SMM_DIR> count-classifications \
--route ask --cycle-id <CLOSE_CYCLE_ID> --since-ts <CLOSE_START_TS>)
<CLOSE_CYCLE_ID> and <CLOSE_START_TS> are emitted by the preload (captured at close-cycle start). --cycle-id excludes concern_classify events tagged with a DIFFERENT close-cycle, so concurrent close-cycles in other teammate worktrees don't leak in (SMM is shared across worktrees). An UNTAGGED event is still counted — the count fails closed rather than silently dropping an ask-route the gate must see — so --since-ts is the bound that keeps pre-cycle events out. Test numerically: [ "$ASK_COUNT" -gt 0 ] → fall through to the shared Step 6 prompt.
No open high-severity concern recorded during Step 4.5, verified via the canonical structured filter:
HIGH_CONCERN_COUNT=$(python3 ${CLAUDE_PLUGIN_ROOT}/smm/smm_cli.py \
--smm-dir <SMM_DIR> count-concerns \
--severity high --cycle-id <CLOSE_CYCLE_ID> --since-ts <CLOSE_START_TS>)
Test numerically: [ "$HIGH_CONCERN_COUNT" -gt 0 ] → fall through to the
shared Step 6 prompt.
The preload emitted a non-empty TEST_COMMAND=... line (sourced from system_context.stack.test_command) AND running that command AFTER all Step 5c fixes landed exits 0. Any non-zero exit means tests aren't green — fall through to the shared Step 6 prompt.
Step 5c classified zero design_decision findings — even if the classifier routed one to fix. Free-close merges into an integration branch (plan or primary), and architectural calls deserve a human checkpoint regardless of the classifier's route choice. Verify via:
DESIGN_DECISION_COUNT=$(python3 ${CLAUDE_PLUGIN_ROOT}/smm/smm_cli.py \
--smm-dir <SMM_DIR> count-classifications \
--category design_decision --cycle-id <CLOSE_CYCLE_ID> --since-ts <CLOSE_START_TS>)
Test numerically: [ "$DESIGN_DECISION_COUNT" -gt 0 ] → fall through to the shared Step 6 prompt.
When TEST_COMMAND is empty (the project hasn't configured a test command), the gate cannot fire. Print this two-line discovery hint before falling through to the shared Step 6 prompt:
Auto-merge disabled — set stack.test_command in system_context.json to enable.
To set it, pipe the command (JSON-quoted) into the edit-stack-field CLI:
printf %s '"<your-test-command>"' | python3 ${CLAUDE_PLUGIN_ROOT}/smm/system_context_cli.py --smm-dir <SMM_DIR> edit-stack-field test_command
When all four conditions hold, print exactly:
"All reviewer findings addressed and tests green — proceeding to merge without confirmation."
then continue to Step 7. Otherwise apply the shared Step 6 AskUserQuestion as written.
python3 ${CLAUDE_PLUGIN_ROOT}/scripts/close_common.py merge \
--cwd . --source <CURRENT_BRANCH> --target <TARGET_BRANCH> \
--smm-dir <SMM_DIR>
Any failing step aborts the chain — source intact for retry. Conflicts are never auto-resolved.
Surface the merge's trailer advisory. The merge step's close_common.py merge prints a Resolves-Event: trailer advisory to stdout when eligible commits fall below the trailer target (fail-open — it never blocks the merge). Tool output is invisible to the user, so if the merge printed that advisory, relay it verbatim — the named commits are still in reach to add trailers.
Tell the user: free branch merged into <TARGET_BRANCH>, PR (if created) merged, local branch deleted. Free-close is complete.
Review the completed sprint. Compares what shipped vs planned, updates execution_plan.json milestones with delivered status, and records sprint velocity. Use when all stories are done or deferred.
Create sprint.json from an execution plan milestone or product spec. Decomposes milestones into context-rich stories with file domains, interface contracts, and inlined design context. Deep codebase dive identifies story boundaries. Customer confirms scope before writing.
The universal per-story execution-shape decision. Reads the session default tier and the plan-reviewer's per-story recommendation, then applies a behavior table: it may exit in-agent (no spawn — continue in the existing checkout) or spawn exactly ONE teammate at the chosen tier (the story's executor_model forwarded to spawn_teammate's --model flag). Valid for solo and parallel frontiers alike. Runs once per story after that story's plan is reviewed.
Push the sprint branch, fork xp-close-reviewer for cross-cutting review, and merge into the target with cleanup. Auto-invoked at the end of /xp-sprint-review; degrades gracefully when gh is unavailable.
Verify acceptance criteria for in-progress stories. Present criteria, guide e2e test execution, mark stories done or deferred, update sprint.json.
This skill should be used when the user needs to select work for the session, review retro Try items, triage open questions, or pick sprint stories. Invoked as part of /xp-kickoff or standalone via /xp-work-selection.