slice-closure
Decide whether the current slice is ready to close, separating slice completion from task completion, and record the result in slice notes and TASK_STATE.md. Use when a slice was implemented, validated or reviewed and may be ready to close. Do not use before slice implementation, during planning, or for overall task progress (use where-we-at). Before closing a long session, use harvest-session-learnings.
- License 许可
- MIT
- 最近更新
- 2026-10-10
- Stars
- ★ 10
- Forks
- 0
01 前言 Front Matter
- name
- slice-closure
- metadata
- { "tools": "Read, Write, Edit, Bash, Glob, Grep", "category": "execution-and-closure", "provenance": "first-party", "x-wos-profiles": "minimal, core, full", "suggested-model": "claude-haiku-4-5", "multi-repo-aware": "true", "primary-cursor-mode": "Ask", "unconditional-loads": "wos/closure-floors.slice-closure.md", "context-layers-consumed": "memory", "context-layers-produced": "memory" }
- description
- Decide whether the current slice is ready to close, separating slice completion from task completion, and record the result in slice notes and TASK_STATE.md. Use when a slice was implemented, validated or reviewed and may be ready to close. Do not use before slice implementation, during planning, or for overall task progress (use where-we-at). Before closing a long session, use harvest-session-learnings.
03 SKILL.md 正文
Output contract, in brief. This body is over the per-skill re-injection cap, so after a compaction the sections below are truncated away while this summary survives. They remain authoritative in full; re-read this file before emitting if you need them.
Standard output layout (required): Produce the command output using this structure (English only):Artifact changes: Follow## Global output contractinWORKFLOW_OPERATING_SYSTEM.mdforAPPLIED/PROPOSED/SKIPrules. Every listed file...Command transcript: Brief audit trail (max 4 lines; max 3 in no-op runs withNO_OP_TRACE).Handoff: Use the adaptive ending format ofWORKFLOW_OPERATING_SYSTEM.md## Global output contract. Every Handoff is one fencedtextb...Definition of done (command output): Closure status is exactly one of: ready / ready-with-followups / not-ready, with evidence.
Act as a senior/staff engineering slice-closure reviewer for the active engineering task.
Goal: Decide whether the current slice is ready to close, without confusing slice completion with full task completion, then persist the result in the task repository as explicit, reviewable closure notes.
Note: this command is opt-in for LOW and MEDIUM complexity slices. Use slice-closure when: work complexity is HIGH, exit criteria require manual verification beyond typecheck, or the slice has follow-ups that need explicit tracking.
Mandatory context bootstrap (before any output):
- Read these sections in
WORKFLOW_OPERATING_SYSTEM.mdfirst:## LLM execution contract## Editor mode policy(mode definitions only; the tool mapping table is lazy-loaded inwos/editor-mode-mappings.mdand needed only for non-Claude-Code tools)## Global output contract(including Adaptive handoff and Mode selection rule)## Cross-cutting workflow guardrails
- Bootstrap tiers: the light-weight commands (
branch-commit,what-next,where-we-at,slice-closure,compact-task-memory) plus the high-frequencyimplement-approved-sliceandsync-task-state(v3 wave1 item D) read the four sections above with two subsections of## Cross-cutting workflow guardrailsskipped:### External web access (centralized)and### Sequencing heuristics (by phase). Everything else is read at every tier, including### Proposal vs approved persistenceand### Substrate peer ownership (per ADR-0034), since all seven write substrate sections and reason about PROPOSED (state-reconcilestays on the full tier for cross-artifact judgment). The full tier is measured at 12109 tokens, the four always-read sections combined; the two skipped subsections are 1,034 of those (measured 2026-09-28), so the reduced tier is about 11,074. The leaf-reviewer tier (verify-against-rubric, ADR-0226) reads only## Global output contract, measured at 5215 tokens, plus its rubric. - Session bootstrap reuse (skip-if-unchanged; v3 wave1 item D): WHEN this conversation already read the bootstrap sections in an earlier turn still VISIBLE in the context window AND
WORKFLOW_OPERATING_SYSTEM.mdhas not changed since, the command MAY skip the re-read and cite the earlier one, emitting one Command transcript line:Bootstrap: reusing turn <N> read, WOS unchanged. Scoped exception to the context-budget re-fetch rule (wos/context-budget.md, "The re-fetch rule"), because these sections are one large, static, byte-identical read repeated every turn; every other tool result still re-fetches. VISIBLE means the section text itself is still present and quotable now, not merely that a record of the earlier read exists: a harness that clears a tool result while the record survives (ADR-0114) has not satisfied VISIBLE, and self-declared memory after a compaction never qualifies. A stateless-per-turn harness is excluded. The transcript line is mandatory; a silent skip is invalid output. - Resolving the spec and
wos/<topic>.md: canonical repository FIRST. Otherwise use active-adapter docs:~/.claude/workflow-docs(CLAUDE_WORKFLOW_DOCS_DIR),~/.cursor/workflow-docs(WORKFLOW_DOCS_DIR; project.cursor/workflow-docson project installs),~/.codex/workflow-docs(CODEX_WORKFLOW_DOCS_DIR), or<kimi-home>/workflow-docs(KIMI_WORKFLOW_DOCS_DIR). Nonempty overrides win. Kimi home:KIMI_CODE_HOME, else existing~/.kimi-code, else existing~/.kimi, default~/.kimi-code. Use one root for both paths; try other adapters if absent or unknown. Report resolved root, fallback to other adapters, or missing mandatory context in### Command transcript. Installed snapshots refresh and prune retired topics only on sync. - Read additional sections only when relevant to this command's role.
- Align all routing recommendations and next-command suggestions with the current command set.
- Official next-command names only: recommendations, including
Run now, use registered names fromcommands/<name>.mdorcommands/<name>/SKILL.md; installed mode uses the available Fhorja command or skill inventory. No invented aliases.Run now: nonewithMode: N/Aends only if no honest next step exists, defined under### Official command names (routing integrity)(ADR-0126). Required inputs: - active task folder path
- TASK_STATE.md
- IMPLEMENTATION_PLAN.md
- current slice artifact, if present
- latest implementation outputs
- latest validation/test results, if available
- relevant real codebase context
- last completed step from TASK_STATE.md (command + summary)
Operating rules:
-
Closure scope boundary. Do not implement code. Never open another command's SKILL or command body from inside this run; never create or edit another slice's SLICES file; never emit a substrate write with an owner other than slice-closure (or its documented co-writer role). Inherited chain authorization ("run the handoff commands without stopping") ends at this run's Handoff: the next command appears ONLY on the
Run now:line, never executed inline. -
Slice-note fill, never rewrite. WHEN the closing slice has a skeleton note with
Status: in progress(born under the slice-note-first floor inimplement-approved-slice), this command FILLS that same file via Edit: files touched, what changed, what was intentionally not changed, validation, residual risks, preserving the skeleton's Goal, Scope and Complexity, and sets itsStatus:to the verdict value; a scope change is recorded as an explicit revision line, never a silent overwrite. -
Handoff: end with the adaptive
### Handoffblock perWORKFLOW_OPERATING_SYSTEM.md## Global output contract(Mode A compact or Mode B full). -
Operating mode (ADR-0008, ADR-0162). WHEN
TASK_STATE.md ## Resume notescontainsOperating mode: minimal,strict,teaching, orassisted, loadwos/operating-modes.mdand apply that mode. WHEN the line is absent, use this command's native rules. Auto-suggestion is not a declaration. -
Substrate write protocol (ADR-0034). MANDATORY for every write to a substrate section this command owns or co-writes (per
wos/substrate-peers.md; slice-closure writes the canonical 5 sections percommands/_shared/task-state-slice-closure-pattern.md). Percommands/_shared/substrate-write-protocol.md ## Concrete computation:sha_beforevia the canonicalsha_of_sectionbash helper (nullonly if the section did not exist before this write).- Header on its own line IMMEDIATELY above the section heading:
<!-- wos:write owner=slice-closure section='## X' run_id=<ULID-or-uuid> ts=<ISO-8601-ms-with-Z> reason=<<=80chars> mode=applied -->. - Write or update the section content.
sha_afterwith the same helper over the post-write section bytes.- Append exactly one JSON line to
active/<task>/.wos/VERIFICATION_LOG.jsonlper the 14-field schema inwos/substrate-peers.md ## Audit trail.sha_afterMUST be valid SHA-256 hex (64 lowercase hex chars), NEVERnullon applied writes;sha_beforeisnullONLY on first write to a fresh section. - slice-closure writes 5 sections in one run per the canonical closure pattern. Repeat steps 1-5 PER section. Reuse the same
run_id+tsacross all 5 section writes in this single invocation.
FORBIDDEN: the half-compliant pattern (JSONL line emitted but inline header omitted, OR
sha_*fields set tonullwhen the section already existed). -
Plan slice Status, by default (P-11). On every run this command SHALL write
Status: closedon the closing slice inIMPLEMENTATION_PLAN.md ## Slices(closed-with-followupswhen every follow-up is a micro-delta;not-readyon that verdict or amaterialfollow-up), whether or not a propagation scope is declared, in ONE call:scripts/emit-substrate-write.sh status --owner slice-closure --file IMPLEMENTATION_PLAN.md --slice <id> --value <value> --task-root <task>edits that line and logs it at## Slices(P-9), with no body file.scripts/check-slice-order.shreads only that line. -
Bounded slice-status propagation (opt-in). DEFAULT = OFF: with no propagation scope declared, this command writes ONLY the canonical 5-section TASK_STATE.md pattern (above) and the plan Status. Propagation fires ONLY when the run declares the bounded scope
propagate-slice-status=<subset of {IMPLEMENTATION_PLAN, SOURCE_OF_TRUTH, README, TEST_STRATEGY}>(via an invocation arg or a per-slicePropagate-status:marker inside### Slice N). For each named target write the bounded slice-status field (closed enum:in-progress | implemented-pending-closure | closed | closed-with-followups | not-ready; verdict "ready to close"->closed, "ready with follow-ups"->closed-with-followups, "not ready"->not-ready) honoring its write regime:- Regime 1 (SUBSTRATE: a
wos:writeheader plus one.wos/VERIFICATION_LOG.jsonlline reusing THIS run's run_id/ts):IMPLEMENTATION_PLAN.mdis the plan Status call above;SOURCE_OF_TRUTH.mdwrites/extends the## Slice statusH2 pointer with a header above that H2 and one JSONL line namingsection='## Slice status'. - Regime 2 (PLAIN -- direct Edit only, NO wos:write header, NO JSONL line):
README.mdandTEST_STRATEGY.md.
- Regime 1 (SUBSTRATE: a
-
Do not reopen broad discovery, broad review, or signed-off contract issues.
-
No-op rule for artifacts: WHEN closure status and evidence are already recorded with no material gap, or the slice documentation or
TASK_STATE.mdwould not materially change, do not rewrite it; return a no-op, route forward (often/sync-task-state), and still output a short NO_OP trace note. The no-op SHALL NOT skip the plan Status write while that line differs from the verdict. -
Treat this as a closure decision for the current slice only, unless the context explicitly says the whole task is ending.
-
First identify the current task scope level and the current closure target.
-
Then evaluate whether the approved slice goal was met. Require the verbatim command output recorded at execution as proof of each exit criterion; an exit criterion without shown evidence is unverified, not met.
-
Closure floors (lazy-loaded). Before emitting any closure verdict, load
wos/closure-floors.slice-closure.mdand apply every floor in it exactly as written there. The floors are: commit-evidence (ADR-0084/0100), experience-verdict (ADR-0091), entry-path probe (ADR-0091), eval-threshold (ADR-0104), integrity (v3 wave3 S1), Layer-2 review, and rollout-constraint reconcile. This load is MANDATORY, not conditional. The closure notes SHALL cite which subsections were read and applied. A slice that does not satisfy a floor takes THAT floor's declared verdict, read from its ownOn missing evidence:line:recordwritesunverified: <reason>into the closing notes and the slice closes,reconcilerecords a named deferral and the slice closes, andrefuseclassifies the slicenot ready to closeand routes per that file (ADR-0203). -
Platform runtime floors (lazy-loaded). WHEN the active task matches the Godot signature (a
project.godotor.gdcodebase, orGODOT_SCENE_PLAN.md/GODOT_RUNTIME_VERIFY.mdin the task folder), the mobile signature (themobile-runtime-targettag, or apackage.jsonwith anexpoorreact-nativedependency plus a generatedandroid/orios/folder), the web signature (theweb-runtime-targettag, or a slice scope touching a servable frontend surface in a project whose manifest declares a web build or preview script), or the backend HTTP signature (thehttp-runtime-targettag, or a slice scope touching an HTTP route handler, controller, or router definition), loadwos/platform-runtime-floors.mdand apply its slice-closure variants (Godot runtime-gate, Godot feel-verdict, Godot tier-declaration, mobile-runtime-gate, web-runtime-gate, backend-runtime-gate) exactly as written there; the closure notes SHALL cite which subsection was read and applied. Inert on a task with no platform signature. -
Pending-verdict checkpoint. WHEN exactly one floor is the only reason a slice is
not ready to close(every other closure input already verified), the closure output SHALL record a fixed-format checkpoint INSIDE the mandatory 5-section TASK_STATE write, zero additional writes:### In progressgainsCheckpoint: awaiting <floor-name> verdict for Slice <N>; investigation complete at commit <ref>; resume via slice-closure with the verdict, and## Recommended next stepnames what produces the verdict. The commit-evidence floor is EXCLUDED. On a later invocation carrying the verdict, resume from the checkpoint WITHOUT re-investigating; confirm only that the cited commit is still the slice's latest; a moved head invalidates the checkpoint and falls back to the full investigation, never a silent stale close. For the human-verdict floors (feel-verdict, experience-verdict) a barePASSargument SHALL NOT resolve the checkpoint: the real artifact block (## Feel verdictor## Experience verdictwithOverall: PASS) must be cited, and the resume is INVALID when this invocation is the same autonomous continuation that recorded the checkpoint under inherited run-without-stopping authorization with no new human turn in between (ADR-0098). Mechanical floors (eval-threshold, the runtime gates) accept the argument verdict. The### Command transcriptSHALL name both the checkpoint write and the checkpoint resume. The checkpoint belongs to THIS command only and never travels into a finalization context:task-closereads the floors themselves. An attended chain on a task branch (ADR-0233) records the floor's ownunverified:line instead. -
Classify the slice into exactly one of:
- ready to close
- ready to close with follow-ups
- not ready to close
-
Distinguish clearly between:
- what was completed
- what is intentionally deferred
- what remains as a blocker inside this slice
-
If the slice is ready, recommend closing it and syncing task state; an attended run sends an uncommitted last slice to
branch-commit --apply(ADR-0233). WHENimplement-fleethanded this HIGH slice over mid-run, the Handoff routes back toimplement-fleetfor the next wave. -
If not, recommend only the smallest remaining action.
-
If status is ready to close with follow-ups, the output SHALL name the gap as
micro-deltaor asmaterial, in those words: explicit micro-deltas under the same slice intent route toimplement-slice-complement; a material gap reopens a fullimplement-approved-slicepass (plan Statusnot-ready). A follow-up left unclassified is incomplete output (ADR-0193). -
Explicitly state what should not be reopened now.
-
Record one complexity and dead-code debt line: did this slice introduce any abstraction, config, dependency, or dead code not yet justified by a current caller or a DECISIONS entry? Default "none"; a non-empty answer becomes a tracked follow-up (## In progress).
-
When proposing a
TASK_STATE.mdupdate, set Work complexity for the next step after closure; definitions inWORKFLOW_OPERATING_SYSTEM.md. -
When adding LEARNINGS during closure, anchor each entry at the exact decision point that failed (
file:line, slice section header, command name, or timestamped TASK_STATE row) pertemplates/LEARNINGS.md## Entry shape; an unanchored entry is disqualified. Add an optionalTags:line (comma-separated keywords) forrank-learnings.sh. -
Multi-repo: when
SOURCE_OF_TRUTH.mdcontains a## Repositoriessection, report exit-criteria validation and remaining blockers per-repo in distinct subsections. A multi-repo slice is "ready to close" only when ALL repos meet their exit criteria; partial-close cases must say which repos passed and which did not.
TASK_STATE.md update pattern (canonical sections to edit at slice closure): Canonical TASK_STATE.md 5-section write pattern. Followed by EVERY command that stamps TASK_STATE.md after a meaningful step (slice-closure, approve-plan, implement-fleet, release-plan, ai-feature-eval-harness, the verify fleets, task-close, and peers). Read "closure" below as "the step this command just completed", then edit exactly these 5 sections in this order:
## Current phase-- if the phase shifted (e.g., discovery -> implementation), update the phase label.## Last completed step-- replace withCommand: <cmd>,Mode: <mode>,Summary: <1-2 line outcome>(the recovery anchor forresume-from-state).### In progress(nested under## Current status, not a standalone H2, pertask-init.md's canonical TASK_STATE.md template) -- if a slice closed cleanly with no follow-up, set to(none). If a follow-up surfaced, list it here as the immediate next item.## Recommended next step-- replace withCommand: <next>,Mode: <mode>,Why: <one line>. Aligns with the HandoffRun nowline.## Current closure target-- if a slice or epic just closed, advance it to the next slice or epic. If the same target still applies, keep it (no-edit OK).
Optional 6th section:
## Resume notes-- update only when external context shifted (a referenced repo moved, a decision was made elsewhere, etc.).
Rules:
- Use Edit (not Write) per section. Avoid full file rewrites.
- If a section would not materially change, skip it (per
WORKFLOW_OPERATING_SYSTEM.md## Cross-cutting workflow guardrails->### Material change (definition)). - All 5 mandatory sections must be present in the file. If any is missing, propose adding it first as a separate edit.
- Total edits per closure typically range 4-6. More than 8 means drift recovery (mark in transcript) or several slices closed at once (split into separate runs).
- Every edit above is a SUBSTRATE WRITE: transaction header plus one
.wos/VERIFICATION_LOG.jsonlline percommands/_shared/substrate-write-protocol.md, most cheaply viascripts/emit-substrate-write.sh. Read that file for the mechanics and for which OTHER targets are in or out; do not infer the boundary from this list. Watch section 3. Four of the five areTASK_STATE.mdH2 sections, but### In progressis an H3 under## Current status, so its write LOGS AT THE OWNING H2 (section='## Current status') and never at### In progress. The validator rejects anysectionnot starting with##(verify-substrate-batch.shfails), andsha_of_sectionruns to the next^##, so hashing the H3 records asha_afterover sibling H3s and surfaces later as false drift at thetask-closeintegrity floor.
Substrate digest fallback
Digest fallback when the canonical helper is unreachable. sha_of_section extracts a section's body with awk and pipes it to shasum -a 256. A run inside a permission boundary that admits shasum but refuses awk and sed cannot invoke that helper, and MUST NOT reimplement it. Assume the helper is unreachable whenever the workflow repository's scripts/ directory is not readable from the working directory.
WHERE the canonical per-section digest helper is unreachable, the write SHALL use a whole-file SHA-256 and SHALL declare the reduced scope:
# One call per file. Read the hash (the first field) out of the output;
# do not pipe it through `cut` or `awk` to trim it.
shasum -a 256 TASK_STATE.md
Then add "sha_scope":"file" to that JSONL line, so a validator distinguishes a file digest from a section digest. The rest of the protocol is unchanged: the transaction header still goes above the section heading, and there is exactly one JSONL line per section write.
Do NOT rebuild the helper by writing each section out as its own file to hash it separately.
At file scope two sections written in the same run share a digest, so the sha_scope field is mandatory, not optional.
Claim grounding (active epistemic humility)
Claim grounding (active epistemic humility). This block governs what you may assert and how you record it. It is keyed to the substrate section you are writing, not to which command is running, and it is INERT on any output that writes none of the claim-bearing sections below. Full contract and rationale: wos/active-epistemic-humility.md.
-
When this applies. This block fires ONLY while you are writing a claim-bearing substrate section:
TASK_STATE.md ## Current known facts,## Risks to watch,## Observations,## Active files in scope,## Canonical decisions;DECISIONS.md ## Locked decisions;IMPLEMENTATION_PLAN.md ## Current gaps,## Risks and mitigations;IMPACT_ANALYSIS.md;EXTERNAL_RESEARCH.md;REFERENCES.md; or any section whose content is a statement a later command or a human decision will act on. WHEN your output writes none of these, this block imposes nothing: skip it and proceed. This is the D-13 inert clause; a fully-grounded or claim-free output pays nothing. -
The unit is the load-bearing claim. A load-bearing claim is one a downstream command or a human decision consumes. A passing aside is not load-bearing; a statement someone will act on is. Apply the rest of this block per load-bearing claim, not per sentence.
-
Ground it or abstain. Before you assert a load-bearing claim, trace it to the enumerable grounded set: a captured
REFERENCES.mdentry, a file read in this session, command output actually seen, or a passing deterministic gate. A claim supported only by model memory is OUTSIDE the grounded set, including when you are right, because that support is not observable. WHEN a load-bearing claim falls outside the set, do NOT assert it: either investigate until it is grounded, or abstain per rule 6. -
Status records provenance, never confidence. WHERE you attach an epistemic status to a claim, the status names WHERE THE CLAIM CAME FROM: a
REFERENCES.mdentry title, a file path plus line, or the gate output it came from. It SHALL NOT express a degree of certainty. Do NOT add a confidence field, a numeric threshold, or a self-assessment prompt anywhere; a self-reported confidence signal is not a usable control signal (wos/active-epistemic-humility.mdPart 1.3). A status whose referent slot is empty is read as UNKNOWN, not as a weak yes. -
Persisted claims carry the status; chat-only claims carry it when they route. Every load-bearing claim you write into a task-memory artifact carries its provenance referent, and that referent travels with the claim so a later command reads it too; do not drop it at the write boundary. A load-bearing claim that appears only in a chat-turn output carries a status only when it crosses the grounding boundary and triggers a route (an abstention, an escalation).
-
Abstain as a routed continuation, never a bare refusal. WHEN you abstain, name the specific investigation that would settle the question AND route to the command that runs it (
capture-references,code-locate,incident-triage, or the fitting one). A withholding that stalls the work is invalid output. Abstention is distinct fromNO_OP:NO_OPmeans there is no work to do; abstention means there is work and the grounding to do it is missing. -
An unfired gate is not evidence. The absence of a fired check does not mean grounding existed. Do not read silence here as a pass.
Standard output layout (required)
Produce the command output using this structure (English only):
Artifact changes
Follow ## Global output contract in WORKFLOW_OPERATING_SYSTEM.md for APPLIED / PROPOSED / SKIP rules. Every listed file carries one of those three tokens, in Lean output too; a prose verb like "written" is not a label.
Command transcript
Brief audit trail (max 4 lines; max 3 in no-op runs with NO_OP_TRACE).
Handoff
Use the adaptive ending format of WORKFLOW_OPERATING_SYSTEM.md ## Global output contract. Every Handoff is one fenced text block with all four lines, Run now:, Mode:, Work complexity: and Reason:, on a stop and on a refusal too; the terminal form is Run now: none with Mode: N/A. A new Mode:, a model or a fresh session is never a stop, an offered choice is one, and Reason: names a role, never a model.
Deliverable status (report, per ADR-0056)
Deliverable reconcile (per ADR-0056). Reconcile the task's ## Requested deliverables ledger in TASK_STATE.md against the delivered work. Hard-fails only when finalizing the whole task; reports without failing at a mid-task checkpoint.
-
Read
## Requested deliverablesinTASK_STATE.md. WHEN the section is absent (legacy task), OR its only row is the- none namedsentinel, this gate is a no-op: skip it and proceed. -
Classify the context. Finalization:
task-close, orreview-hardas the pre-PR final pass. Checkpoint:where-we-at,slice-closure, and any otherreview-hardrun. At a checkpoint anin-scoperow not yet done is normal remaining work. -
Reconciled vs silent omission. A row is reconciled when it is
doneorde-scoped:<reason>with that reason recorded inDECISIONS.md. A deliverable named in the brief with NO ledger row, or a row dropped without a recorded de-scope, is a silent omission. You MUST cross-check the ledger against the brief: read the task'sREADME.md(whichtask-initseeds from the brief, its## Briefsection) and the original request when in context, and confirm every deliverable named there has a row. WHEN no brief artifact is available, reconcile the rows that exist and state that ledger-vs-brief completeness could not be re-verified (do not claim it was). -
Apply the gate by context.
- WHEN finalizing: IF any row is unreconciled (still
in-scope, or a silent omission per step 3), THEN this command's output is invalid. Name each unreconciled deliverable, say whether it should be delivered or de-scoped, and route toimplementation-plan(plan the missing work) ordecision-interview(a de-scope, which only the person's answer records). A provisional decision never de-scopes (ADR-0233): in an attended chain on a task branch or under standing authorization (ADR-0244) the row staysin-scopeand is listed asNot delivered, needs you: <deliverable>: <what is undecided>inTASK_STATE.md ## Open questions / blockers:task-closestill holds it unreconciled,review-hard's pre-PR pass only lists it. - WHILE at a checkpoint: report each not-yet-done
in-scoperow as remaining work and do NOT invalidate output on that basis. A silent omission (step 3) is NOT normal progress: name the missing deliverable, record it in theTASK_STATE.mdcheckpoint output as a must-address finding (never a bare one-line mention), and route it as the finalization branch does. Neither case invalidates the output at a checkpoint.
- WHEN finalizing: IF any row is unreconciled (still
A de-scope is allowed; silence is not.
References status (report, X2)
References reconcile. Reconcile the references a task cited (REFERENCES.md, EXTERNAL_RESEARCH.md, or project-level references) against what it shipped. Reports at a mid-task checkpoint; hard-fails only when finalizing the whole task.
-
This sub-check fires only WHEN the task produced or cited references: a
REFERENCES.mdorEXTERNAL_RESEARCH.mdin the task folder, or aGrounded in:citation in the shipped work. WHEN none is present, it is a no-op: skip and proceed. -
Finalization:
task-close(orreview-hardas the pre-PR final pass). Checkpoint:slice-closureorwhere-we-at. At a checkpoint a cited reference not yet reflected is normal in-progress work. -
Reconcile cited vs reflected. For each cited reference, confirm the shipped work materially reflects it (a real layout, behavior, or decision traceable to it), not merely a name-drop. A reference with no material trace is cited-but-unused.
-
Apply the gate by context.
- WHEN finalizing: IF any cited reference is unused (no material trace) THEN name it and require removing the citation or pointing to where it is reflected, and route to
implement-slice-complementbefore closing. - WHILE at a checkpoint: report each cited-but-unused reference as a must-address finding (name it, route to
implement-slice-complement), and do NOT invalidate the whole output on that basis.
- WHEN finalizing: IF any cited reference is unused (no material trace) THEN name it and require removing the citation or pointing to where it is reflected, and route to
"Cite only what you used" is the invariant.
Definition of done (command output)
- Closure status is exactly one of: ready / ready-with-followups / not-ready, with evidence.
- Commit-evidence floor (ADR-0084, ADR-0100): a ready-to-close slice cites its commit reference or records an explicit committing-waiver covering only genuinely discardable work; real work pending a human commit is a bounded deferral that keeps the slice open (not a waiver), and a slice with none of the three is classified not-ready and routed to
branch-commit. - Eval-threshold floor (ADR-0104): when an
AI_EVAL_PLAN.mdcovers the closing slice's feature, closure requires the cited score-vs-threshold outcome met on the held-out set (or an explicit bounded skip reason per ADR-0098); harness-mechanism wording never substitutes for the threshold outcome. - Closure floors (lazy, unconditional;
wos/closure-floors.slice-closure.md): the slice-closure variants were loaded and applied, and the closure notes cite which subsections were read. Emitting a closure verdict without that citation is invalid output. - Layer-2 review floor: a cited
review-hardverdict, inline or dispatched (plussecurity-reviewon a security surface), a host-repository gate cited by an artifact it produced, or an explicit one-line skip reason; a review that followed no Fhorja contract does not count (ADR-0244). Layer 1 green never substitutes. - Rollout-constraint reconcile: when the plan carries
## Rollout and rollback notes, every constraint naming a file, config key, or release track inside the closing slice's Scope is satisfied or recorded as a named deferral. - Platform runtime floors: on a Godot-, mobile-, web-, or backend-HTTP-signature task, the slice-closure variants in
wos/platform-runtime-floors.mdwere applied and the closure notes cite which subsection was read; a slice failing a floor takes the verdict itsOn missing evidence:line declares and routes per that file. Inert on tasks with no platform signature. - Experience gates (
On missing evidence: record): a slice taggeduser-facing-contentornew-user-facing-surfacerecordsunverified: no experience verdict on a sample, and anew-user-facing-surfacesliceunverified: entry path never exercised, unless the verdict or the entry-path run is cited or a skip reason is recorded (ADR-0203); stands down on the Godot signature in favor of the D-4 floor above. - Clearly distinguishes slice completion vs full task completion.
- Slice file updates are
APPLIED. - Integrity floor (v3 wave3, S1): verify-substrate-batch ran and exited 0, or the explicit
integrity-waiver: N advisories unresolved (<reason>)line is recorded; a silent non-zero closure is invalid output. - Before declaring this output done, confirm it satisfies the shared Definition of done (command outputs) and Gate conditions in WORKFLOW_OPERATING_SYSTEM.md.
Quality bar: Optimize for clean closure, scope discipline, low ambiguity, and forward momentum.
04 同仓库更多 Skills
fixture-status
Fixture skill that reads status only. Use when you want a quick status. Do not use when several plans run at once (use fixture-plan-fleet).
fixture-plan
Fixture skill that plans a thing end to end. Use when the plan is missing. Do not use when you only need a status read.
fixture-plan-fleet
Fixture skill, the fleet variant of fixture-plan. Use when several plans run at once. Do not use for a single plan (use fixture-plan).
fixture-status
Fixture skill that reads status only. Use when you want a quick status. Do not use when several plans run at once (use fixture-plan-fleet).
fixture-plan
Fixture skill that plans a thing end to end. Use when the plan is missing. Do not use when you only need a status read (use fixture-status).
fixture-plan-fleet
Fixture skill, the fleet variant of fixture-plan. Use when several plans run at once. Do not use for a single plan (use fixture-plan).