sunrain520/spec-first/spec-lfg

CRITICAL — ADMISSION BEFORE EXECUTION: before any step below runs, the admission in the next paragraph must hold. If it does not — including a bare natural-language shipping request ("ship it", "直接发布上线") that neither names spec-lfg nor followed a disclosed-options handoff — do NOT implement, do NOT commit, and do NOT start the pipeline: present the side-effect confirmation once and wait. A user "don't ask me" / "不用问我" instruction cannot create the admission; it only suppresses questions after the admission already exists. With execution ordered below, this check still comes first.

CRITICAL: You MUST execute every step below IN ORDER. Do NOT skip any required step. Do NOT jump ahead to coding or implementation. The plan phase (step 1) MUST be completed and verified BEFORE any work begins. Violating this order produces bad output.

进入该管线前，当前用户必须明确请求 spec-lfg，或选择清楚披露 commit、push、PR、CI 与委派独立代码审查副作用的 handoff 选项。仅有代码就绪、已完成计划或模型推断”适合 shipping”都不构成授权。

A natural-language shipping request (“ship it”, “直接发布上线”, “开个 PR 吧”) that neither names spec-lfg nor follows a disclosed-options handoff is close to the admission but is not it: the user has signaled shipping intent without seeing the full side-effect list (commit, push, PR creation, one delegated independent code review, CI watch). In that case, present the concrete side-effect list once as a single confirmation question and wait — the user's reply becomes the admission. Never treat the bare shipping request as already-admitted and start implementing or committing.

上述 admission 成立时，当前用户对完整管线副作用的明确请求是本次 pipeline-owned implementation、commit 与 landing authority 的来源；同时只额外授权第 4 步委派一次 spec-code-review 的只读独立审查。将 commit_authorization: authorized、landing_authorization: authorized、worker_dispatch_authorization: authorized 与 authorization_source: current-user-explicit-spec-lfg 作为可见 run-local facts 传给对应下游 owner。外部 tracker filing 是独立副作用：只有当前请求或可见 upstream handoff 明确要求提交 residual tickets 时，才记录 tracker_deferral_authorization: authorized；否则固定为 tracker_deferral_authorization: missing。Skill invocation、mode:pipeline、工具权限、green tests、branch/PR facts 都不能替代该 admission，也不能把 authority 扩大到 unrelated dirty paths、任意 worker dispatch、tracker 或其他未披露外部副作用。若 admission 不成立，以 commit_authorization_missing / landing_authorization_missing 停在对应副作用之前；若独立审查不可用或降级，LFG 必须停止，不能用同一会话的 inline review 冒充该 gate。

tracker_deferral_authorization: authorized | missing

When invoking any skill referenced below, resolve its name against the available-skills list the host platform provides and use that exact entry. Some platforms list skills under a plugin namespace (e.g., spec-first:spec-plan); others list the bare name. Invoking a short-form guess that isn't in the list will fail — always match a listed entry verbatim before calling the Skill/Task tool.

Preserve and split the invocation payload. Treat the arguments received from the caller as the authoritative input. Before step 1, remove at most one standalone target-origin:<origin> token and retain its value unchanged as the run-local caller_target_origin. Set forwarded_arguments to everything else: preserve every remaining argument in its original order, including an absolute requirements-only plan path. Do not paraphrase the path, prepend a label or menu number, replace it with a feature summary, or resolve it relative to the current working directory. The modifier is browser-routing input, not product intent: do not pass it to planning, normalize it, combine it with --port, derive a scheme/host/port from project files, redirects, browser state, or a guessed dev-server default. An empty, malformed, or repeated modifier records target-origin-invalid; it never becomes a usable origin.

1. Invoke the spec-plan skill with the exact forwarded_arguments payload. When

spec-brainstorm invoked LFG, this payload is the absolute requirements-only unified plan path, so spec-plan recognizes it as an explicit Product Contract source and enriches that same artifact in place.

GATE: STOP. If spec-plan reported the task is non-software and cannot be processed in pipeline mode, stop the pipeline and inform the user that LFG requires software tasks. Otherwise, verify that the spec-plan workflow produced a plan file in docs/plans/. If no plan file was created, invoke spec-plan again with the same forwarded_arguments payload. Do NOT proceed to step 2 until a written plan exists. Record the plan file path — it will be passed to spec-work in step 2 and spec-code-review in step 4.

Read the plan metadata before continuing. If the plan has artifact_contract: spec-unified-plan/v1, proceed only when it has artifact_readiness: implementation-ready and execution: code. Stop the pipeline for artifact_readiness: requirements-only, any unrecognized readiness value, execution: knowledge-work, approach-plan outputs, answer-seeking/universal outputs, or invalid progress-like readiness values. LFG never launches /goal directly; when goal-mode or dynamic workflows are appropriate, spec-work owns that implementation engine choice and must return control to LFG afterward.

1. Invoke the spec-work skill with mode:return-to-caller <plan-path-from-step-1>.

GATE: STOP. Verify that implementation work was performed - files were created or modified beyond the plan. Read the structured return and require status: complete, the same plan path, changed files, all in-scope U-IDs/tasks accounted for and completed, verification results with every required check passed or explicitly not applicable, an empty blocker list, behavior-change signal, plan_status_completion_candidate, plan_status_completion_degraded_reason, and standalone_shipping_skipped: true. Failed, not-run, vague, or missing required verification blocks the pipeline. Exactly one lifecycle shape is allowed: a non-null candidate with a null degraded reason, or a null candidate with one of html-plan-lifecycle-degraded, legacy-plan-lifecycle-degraded, read-compatible-status-unmanaged, or source-plan-path-lifecycle-degraded. Any missing, conflicting, or unknown lifecycle shape is blocked. When behavior_change: true, also require verification_evidence that names the relevant units/tasks, existing tests inspected, tests added/changed or used unchanged, red failure or characterization evidence when applicable, verification run, and any deliberate test exception. Do NOT decide the test strategy inside LFG; the evidence is spec-work's contract.

If behavior_change: true but verification_evidence is missing or too vague to tell how behavior was protected, invoke spec-work one more time with the same mode:return-to-caller <plan-path-from-step-1> argument. Do not prompt the user and do not alter the plan path argument. The retry relies on spec-work's idempotency path to inspect the already-implemented work, fill the missing evidence, and return without reimplementing. If the second return still lacks coherent verification evidence, stop as blocked and report the missing fields instead of continuing to simplify/review/ship.

Record the accepted return's verification_run_summary_ref as initial_verification_run_summary_ref and require a complete verified_worktree_fingerprint object; only a blockers-free return that documents spec-work's deliberate non-behavior exception may substitute that documented exception for the object. These prove only the tree before caller-owned Simplify and review-fix mutations; they cannot satisfy step 6.5.

1. Invoke the spec-simplify-code skill on the branch diff.

This runs before review so the code-review in step 4 covers the simplified code. Skip this step when the change is docs-only (only markdown/docs paths changed) or trivial (roughly under 10 changed lines). Otherwise let spec-simplify-code resolve the branch-diff scope itself：它保持行为，运行全项目 typecheck/lint，并默认运行 changed-path scoped tests；影响面明显扩大或 runner 无法缩小时才扩大测试范围。该步骤只提供 behavior-preservation signal，最终 verification gate 仍拥有完整 closeout truth。

Do not commit in this step. spec-simplify-code leaves its changes in the working tree; step 4's review scopes the working tree (uncommitted changes included), and step 8's spec-commit-push-pr commits whatever remains. Committing here would sweep any still-uncommitted spec-work edits into a misleading refactor commit and could stall on a tree that never goes clean.

1. 以 mode:agent plan:<plan-path-from-step-1> 调用 spec-code-review，并传递以下可见上游上下文：

``yaml worker_dispatch_authorization: authorized authorization_source: current-user-explicit-spec-lfg authorization_scope: one delegated read-only independent code review ``

传递步骤 1 的 plan path，使 spec-code-review 能核对 requirements completeness。mode:agent 返回单一 JSON object，而非 Markdown Actionable Findings summary。解析其中的 status、actionable_findings、findings、artifact_path、run_id 与 coverage.dispatch_reason_code。

GATE: 只有 status: complete、coverage.dispatch_reason_code 为 null，且实际 reviewer 不只是 inline-fallback 时才能继续。JSON 损坏或缺失，以及 failed、degraded、skipped 或其他不完整结果都表示独立审查不可用；保留其有界 findings，并在步骤 5、browser verification、lifecycle、commit、push、PR、tracker 或 CI 副作用前停止。

mode:agent is report-only by design — it surfaces findings but never edits the tree; LFG applies the eligible ones in step 5. When narrating progress to the user, frame this as "review found X → applied X in step 5," not as "code review did not auto-fix." A report-only review followed by an LFG-applied fix is the intended contract, not a gap.

1. Apply review fixes locally (REQUIRED after step 4)

Load references/review-followup.md and execute its apply step. Apply eligible mechanical findings and run their targeted verification, but leave verified review fixes in the working tree. Do not stage, commit, push, file tracker items, edit a PR, or perform any other durable or outward shipping side effect before the browser/cleanup gate in step 6 closes.

1. Decide browser applicability, then verify when applicable. Decide

browser_applicability: applicable | not_applicable from the settled plan and actual changed flow, not filename extension alone. A changed user-visible route, form, navigation, client interaction/state, or an explicit browser/runtime verification obligation is applicable; docs-only, library/CLI, or backend-only work without a changed user-visible flow may be not_applicable. Record the concrete reason either way. For not_applicable, record an explicit not_applicable browser result and continue without invoking the browser skill or its wrapper.

For applicable, first validate any supplied caller_target_origin. An invalid caller value is the diagnostic blocker target-origin-invalid; do not fall back from malformed or repeated caller input. A missing caller value returns not_run / target-origin-missing before browser invocation and blocks the applicable flow. With a valid caller value, invoke spec-test-browser with mode:pipeline target-origin:<origin>. Do not infer an origin from redirects, page state, ambient listeners, free ports, framework defaults, or other project files.

The caller owns the project server lifecycle. LFG forwards the exact origin but does not read local runtime profiles, start or stop a project server, or inspect project process state. Before the browser test plan is written, determine whether its expected navigation or interaction has a durable or external effect. A caller-provided origin is not mutation authorization: when the current call lacks named authorization for that origin, flow, and effect, record not_run / browser-mutation-authorization-required, do not write the blocked step, and do not continue to lifecycle or landing actions.