review: admission PASS — light-kanban-worker admitted via review-loop agent-skill

Charter rev 1, round-01: F-001/F-002/F-003 resolved, F-004 rejected, fresh
Evaluator PASS criterion-by-criterion. Catalog and guides aligned with the
post-admission pre-release state.

Author: Light Cheng

CATALOG.md

@@ -61,7 +61,7 @@ dependencies and modified third-party variants are documented separately.
 - **Purpose:** Pick up and execute one Light-Kanban task per scheduled agent run, then return it for human confirmation; resumes owned in-progress work and review feedback before claiming new tasks.
 - **Invocation:** Model-invoked; manual entry point is supported.
 - **Package:** [skills/light-kanban-worker/](skills/light-kanban-worker/)
-- **Status:** Admitted first-party through the full admission path (`review-loop agent-skill`); released in v0.1.4.
+- **Status:** Admitted first-party through the full admission path (`review-loop agent-skill` `PASS`); scheduled for the v0.1.4 release.
 - **Evidence:** Contract and behavior tests under [skills/light-kanban-worker/tests/](skills/light-kanban-worker/tests/), [user guide](docs/skills/light-kanban-worker.md), and [admission evidence](docs/evidence/admissions/light-kanban-worker/README.md).
 - **Installation path:** `skills/light-kanban-worker/` in a host-recognized Skills root.
 

CATALOG.zh-CN.md

@@ -54,7 +54,7 @@
 - **作用：** 每次定时运行领取并执行一张 Light-Kanban 任务，交回人工验收；先继续自己持有的任务和 review feedback，再领取新任务。
 - **调用：** Model-invoked，支持手动入口。
 - **包：** [skills/light-kanban-worker/](skills/light-kanban-worker/)
-- **状态：** 通过完整准入路径（`review-loop agent-skill`）准入；v0.1.4 中发布。
+- **状态：** 通过完整准入路径（`review-loop agent-skill` `PASS`）准入；计划随 v0.1.4 发布。
 - **证据：** [contract 与 behavior 测试](skills/light-kanban-worker/tests/)、[使用指南](docs/zh-CN/skills/light-kanban-worker.md)与[准入证据](docs/evidence/admissions/light-kanban-worker/README.zh-CN.md)。
 - **安装路径：** host 认可的 Skills root 下的 `skills/light-kanban-worker/`。
 

docs/evidence/admissions/light-kanban-worker/README.md

@@ -8,30 +8,23 @@
 - Invocation type: model-invoked, with a supported manual entry point
 - Profile: `review-loop` `agent-skill`
 - Stable-release boundary: v0.1.3 contains seven packages and does not contain `light-kanban-worker`
-- Admission status: `IN PROGRESS` — full admission path; the prompt-only fast track does not apply because the Skill accesses the network, reads workspace files, and mutates Light-Kanban task state
-
-## Evidence plan (full path)
-
-| Area | Required demonstration |
-| --- | --- |
-| Structure | Package tree, `SKILL.md` metadata, internal links and resources validate with the package contract and behavior suites. |
-| Installation and discovery | A fresh environment installs the package form, then discovers the installed Skill without relying on the source checkout. |
-| Behavioral | Scenarios A–F against a real Light-Kanban server: fresh task, Request Changes rework, two-worker atomic claim, workspace missing → block, empty queue → no mutation, offline → no mutation with clear failure. |
-| Invocation | A scheduled-style prompt and a one-shot manual prompt both resolve to the worker; model-invoked metadata agrees with `SKILL.md`. |
-| Review | `review-loop` with the `agent-skill` Profile evaluates the candidate using Producer evidence and a fresh Evaluator. |
-| Attribution | Owner-authored first-party; no third-party content, so no `ATTRIBUTION.md` is required. |
-
-## Results
-
-The rows above are filled in with exact commands, environment facts, inputs,
-outputs, and limitations as each gate completes. The final verdict is owned
-by `review-loop agent-skill`; only a `PASS` admits the package into the
-first-party collection.
-
-- Contract and behavior suites:
-  `skills/light-kanban-worker/tests/` (positive and negative fixtures,
-  non-zero assertions).
-- Behavioral evidence against a real Light-Kanban server: recorded in
-  [behavioral-evidence.md](behavioral-evidence.md).
-- Final acceptance: recorded under [review-loop/](review-loop/) once the
-  `review-loop agent-skill` run completes.
+- Admission status: `PASS` — full admission path (2026-08-16); the prompt-only fast track does not apply because the Skill accesses the network, reads workspace files, and mutates Light-Kanban task state
+
+## Evidence summary
+
+| Area | Result | Evidence boundary |
+| --- | --- | --- |
+| Structure | PASS | Package contract and behavior suites (metadata, required workflow sections, golden-flow order, rule checkers) plus collection discovery/contract composition; 90 collection assertions + all 19 package suites green. |
+| Installation and discovery | PASS | Clean-copy install to `/tmp/lk-worker-fresh/codex/skills/light-kanban-worker`: 10/10 files SHA-256 identical, tree contains only the declared package, both suites green against the installed copy (shared collection test harness on PYTHONPATH). Published-tag `npx skills add` verification is the v0.1.4 release gate. |
+| Behavioral | PASS | Scenarios A–F against a real Light-Kanban server (binary built from `LightDevCoder/light-kanban` main, commit `f49ace5`): fresh task, Request Changes rework, two-worker atomic claim (exactly one 200 / one 409), workspace missing → block, empty queue no-mutation, offline no-mutation. Full transcript recorded. |
+| Invocation | PASS | Fresh read-only probes: the scheduled-task prompt loads the skill with the correct first protocol actions; an unrelated instruction does not trigger it; the skill invokes no other user-invoked skill. |
+| Review | PASS | `review-loop` `agent-skill` Profile, Charter revision 1: Critic candidates F-001/F-002/F-003 resolved with verified repairs, F-004 rejected; fresh independent Evaluator returned `PASS` criterion-by-criterion. |
+| Attribution | PASS | Owner-authored first-party; no third-party content or code, so no `ATTRIBUTION.md` is required. |
+
+Behavioral evidence: [behavioral-evidence.md](behavioral-evidence.md).
+Final acceptance record: [review-loop/](review-loop/) (charter, findings,
+round-01 records, verdict).
+
+Admission does not waive release, installation-command verification, or
+Program-level acceptance gates; those are recorded under
+[docs/evidence/releases/v0.1.4/](../../releases/v0.1.4/).

docs/evidence/admissions/light-kanban-worker/README.zh-CN.md

@@ -8,26 +8,21 @@
 - 调用类型：model-invoked，支持手动入口
 - Profile：`review-loop` `agent-skill`
 - 稳定版本边界：v0.1.3 包含七个包，不含 `light-kanban-worker`
-- 准入状态：`IN PROGRESS` — 完整准入路径；该 Skill 访问网络、读取 workspace 文件、改变 Light-Kanban 任务状态，纯提示型快速通道不适用
+- 准入状态：`PASS` — 完整准入路径（2026-08-16）；该 Skill 访问网络、读取 workspace 文件、改变 Light-Kanban 任务状态，纯提示型快速通道不适用
 
-## 证据计划（完整路径）
+## 证据摘要
 
-| 领域 | 需要证明的内容 |
-| --- | --- |
-| 结构 | 包结构、`SKILL.md` metadata、内部链接与资源通过包级 contract 与 behavior 套件。 |
-| 安装与发现 | 全新环境安装该包，并在不依赖 source checkout 的情况下发现已安装 Skill。 |
-| 行为 | 针对真实 Light-Kanban 服务器运行场景 A–F：新任务、Request Changes 返工、双 worker 原子 claim、workspace 缺失 → block、空队列 → 无变更、离线 → 无变更且失败清晰。 |
-| 调用 | scheduled 风格 prompt 与一次性手动 prompt 都能命中 worker；model-invoked metadata 与 `SKILL.md` 一致。 |
-| 审查 | `review-loop` 以 `agent-skill` Profile 评估候选，使用 Producer evidence 与全新 Evaluator。 |
-| 归属 | 第一方原创；无第三方内容，无需 `ATTRIBUTION.md`。 |
+| 领域 | 结果 | 证据边界 |
+| --- | --- | --- |
+| 结构 | PASS | 包级 contract 与 behavior 套件（metadata、必需 workflow 章节、golden-flow 顺序、规则 checker）加 collection discovery/contract 组合；90 条 collection 断言与全部 19 个包套件通过。 |
+| 安装与发现 | PASS | 干净复制安装至 `/tmp/lk-worker-fresh/codex/skills/light-kanban-worker`：10/10 文件 SHA-256 一致，安装树仅含声明的包内容，两个套件在安装副本上通过（共享 collection 测试 harness 在 PYTHONPATH 上）。已发布 tag 的 `npx skills add` 验证是 v0.1.4 release gate。 |
+| 行为 | PASS | 针对真实 Light-Kanban 服务器（由 `LightDevCoder/light-kanban` main，commit `f49ace5` 构建）的场景 A–F：新任务、Request Changes 返工、双 worker 原子 claim（恰好一个 200 / 一个 409）、workspace 缺失 → block、空队列无变更、离线无变更。完整 transcript 已记录。 |
+| 调用 | PASS | 全新只读探针：scheduled-task prompt 以正确的首步协议加载该 Skill；无关指令不触发；该 Skill 不调用任何其他 user-invoked Skill。 |
+| 审查 | PASS | `review-loop` `agent-skill` Profile，Charter revision 1：Critic 候选 F-001/F-002/F-003 已修复解决，F-004 驳回；全新独立 Evaluator 逐条给出 `PASS`。 |
+| 归属 | PASS | 第一方原创；无第三方内容或代码，无需 `ATTRIBUTION.md`。 |
 
-## 结果
+行为证据：[behavioral-evidence.zh-CN.md](behavioral-evidence.zh-CN.md)。
+最终验收记录：[review-loop/](review-loop/)（charter、findings、round-01 记录、verdict）。
 
-每个 gate 完成后，用确切命令、环境事实、输入、输出和限制填写上表。最终裁决
-由 `review-loop agent-skill` 拥有；只有 `PASS` 才能进入第一方集合。
-
-- contract 与 behavior 套件：`skills/light-kanban-worker/tests/`
-  （positive/negative fixtures、非零断言）。
-- 真实 Light-Kanban 服务器的行为证据：见
-  [behavioral-evidence.md](behavioral-evidence.md)。
-- 最终验收：`review-loop agent-skill` 运行完成后记录在 [review-loop/](review-loop/)。
+准入不免除 release、安装命令验证或 Program 级验收 gate；这些记录在
+[docs/evidence/releases/v0.1.4/](../../releases/v0.1.4/)。

docs/evidence/admissions/light-kanban-worker/review-loop/README.md

@@ -1,9 +1,22 @@
 # `light-kanban-worker` — `review-loop agent-skill` acceptance record
 
-Status: `PENDING` — the full-path review run starts after the structural,
-behavioral, and invocation evidence is collected. This directory holds the
-frozen acceptance source, round evidence, findings, repairs, and the final
-`PASS` / `FAIL` / `BLOCKED` verdict owned by `review-loop` with the
-`agent-skill` Profile.
+Status: `PASS` — final acceptance by `review-loop` with the `agent-skill`
+Profile (Charter revision 1, round-01, 2026-08-16). Independence: full
+(separate fresh read-only Critic and Evaluator contexts).
 
-Only a final `PASS` admits the package into the first-party collection.
+## Record contents
+
+- [charter.md](charter.md) — frozen acceptance baseline (SPEC sections 4–35,
+  57–58, 69; AC-1…AC-14)
+- [findings.md](findings.md) — canonical registry: F-001/F-002/F-003 resolved,
+  F-004 rejected
+- [verdict.md](verdict.md) — final `PASS` and reopen note
+- [rounds/round-01/](rounds/round-01/) — producer evidence, critic findings,
+  repair evidence, evaluator verdict
+
+## Verdict
+
+`PASS`: the package satisfies all fourteen frozen acceptance criteria with
+appropriately labeled structural, installation, behavioral, invocation, and
+review evidence; the three confirmed findings were repaired in scope and
+verified green by the fresh Evaluator.

docs/evidence/admissions/light-kanban-worker/review-loop/README.zh-CN.md

@@ -1,7 +1,18 @@
 # `light-kanban-worker` — `review-loop agent-skill` 验收记录
 
-状态：`PENDING` — 结构、行为与调用证据收集完成后启动完整路径审查。本目录保存
-冻结的 acceptance source、各轮证据、findings、修复记录，以及 `review-loop`
-以 `agent-skill` Profile 拥有的最终 `PASS` / `FAIL` / `BLOCKED` verdict。
+状态：`PASS` — `review-loop` 以 `agent-skill` Profile 完成最终验收（Charter
+revision 1，round-01，2026-08-16）。独立性：full（Critic 与 Evaluator 为两个
+独立的全新只读上下文）。
 
-只有最终 `PASS` 才能让该包进入第一方集合。
+## 记录内容
+
+- [charter.md](charter.md) — 冻结的验收基线（SPEC 第 4–35、57–58、69 节；AC-1…AC-14）
+- [findings.md](findings.md) — 规范登记：F-001/F-002/F-003 已解决，F-004 已驳回
+- [verdict.md](verdict.md) — 最终 `PASS` 与重新开启说明
+- [rounds/round-01/](rounds/round-01/) — producer evidence、critic findings、repair evidence、evaluator verdict
+
+## Verdict
+
+`PASS`：该包满足全部十四项冻结验收标准，证据标签准确（structural /
+installation / behavioral / invocation / review）；三项确认 findings 已在
+范围内修复，并由全新 Evaluator 验证通过。

docs/evidence/admissions/light-kanban-worker/review-loop/rounds/round-01/evaluator-verdict.md

@@ -0,0 +1,44 @@
+# Evaluator Verdict - Round 1
+
+- Context identity: fresh read-only subagent (separate context from the Critic)
+- Declared independence: full (read-only; no modifications to target, charter, state, or evidence; re-ran suites independently)
+- Charter revision: 1
+- Profile: agent-skill
+
+## Criterion-by-criterion judgment
+
+| Criterion | Judgment | Key evidence (label) |
+| --- | --- | --- |
+| AC-1 structure & metadata | satisfied | contract/behavior suites (structural); 10-file clean copy byte-identical (installation) |
+| AC-2 model-invoked + manual + scheduled | satisfied | no disable-model-invocation, allow_implicit_invocation true (structural); trigger/non-trigger probes (invocation) |
+| AC-3 stable identity, reuse, no guessing | satisfied | F-003 repair verified (structural); scenarios A first registration / B reuse (behavioral) |
+| AC-4 existing-work-first, reviewFeedback priority, one task | satisfied | ordering + priority checkers (structural); scenario B rework chain (behavioral) |
+| AC-5 atomic claim, bounded retry ≤2 | satisfied | SKILL.md/api.md (structural); scenario C exactly one 200 / one 409 (behavioral) |
+| AC-6 no work → clean exit | satisfied | "No task available" section (structural); scenario E db SHA-1 unchanged (behavioral) |
+| AC-7 workspace validation → block | satisfied | exact reason pinned (structural); scenario D blocked with reason (behavioral) |
+| AC-8 complete → stop; human-only boundary; failure handling | satisfied | F-001 repaired canonical boundary (structural); A/B/D/F (behavioral) |
+| AC-9 no daemon/polling/runtime scripts | satisfied | no-daemon checker + 10-file tree has no executables (structural); daemon fixture rejected (negative) |
+| AC-10 API reference coverage | satisfied | seven endpoints + field explanations (structural); source cross-check (source) |
+| AC-11 tests with positive/negative fixtures, non-zero assertions | satisfied | suites green, assertGreater(assertions, 0), F-002 single-rule precision (structural) |
+| AC-12 scenarios A–F vs real server | satisfied | behavioral-evidence.md, all PASS (behavioral) |
+| AC-13 review-loop agent-skill acceptance | satisfied | this loop (review) |
+| AC-14 v1.0.4+ compatibility, no new API | satisfied | SKILL.md/api.md statement + E-007 no endpoint changes (source) |
+
+## Open findings and approved exceptions
+- F-001, F-002, F-003: resolved with verified repair evidence.
+- F-004: rejected with recorded rationale (fields used by ordering rules).
+- No approved exceptions in the Charter.
+
+## Residual concerns (Low, non-blocking, already recorded Charter limitations)
+1. AC-3 identity enforcement is instruction-level (no static pin of the two sources).
+2. AC-12 LAN cross-machine reachability is documented but not live-tested.
+
+## Verdict
+PASS
+
+Reason: the package satisfies AC-1 through AC-14, the three repairs are
+correctly applied with a green self-run suite, and the only residual items are
+low-severity and within the Charter's recorded limitations.
+
+Next action: record the verdict, update the admission record to PASS, and
+proceed to the v0.1.4 release gates (out of admission scope).

docs/evidence/admissions/light-kanban-worker/review-loop/state.md

@@ -4,10 +4,10 @@
 - Goal: admit `light-kanban-worker` into the first-party collection (full path)
 - Charter revision: 1 (approved — direct user SPEC)
 - Profile: agent-skill
-- Status: READY
-- Current round: round-01 (in progress)
-- Independence requirement: separate fresh read-only Critic and Evaluator contexts (subagents)
+- Status: CLOSED — PASS (round-01)
+- Current round: round-01 (closed)
+- Independence: full (separate fresh read-only Critic and Evaluator subagent contexts)
 - Blockers: none
-- Recent decisions: charter frozen; clean-copy installation evidence collected; behavioral scenarios A–F recorded PASS; invocation trigger/non-trigger probes completed
+- Recent decisions: verdict PASS; F-001/F-002/F-003 resolved, F-004 rejected; repairs verified by the fresh Evaluator
 - Open questions: none
-- Next action: launch read-only Critic with the frozen charter, bounded target view, admissible evidence, and the finding schema; then validate candidates; then a fresh Evaluator; then apply the verdict
+- Next action: v0.1.4 release gates (tag, fresh-install verification, release receipt) — out of admission scope

docs/evidence/admissions/light-kanban-worker/review-loop/verdict.md

@@ -0,0 +1,42 @@
+# Verdict
+
+- Charter revision: 1
+- Profile: agent-skill
+- Verdict: **PASS**
+- Round: round-01
+- Independence: full (separate fresh read-only Critic and Evaluator contexts)
+- Date: 2026-08-16
+
+## Conclusion
+
+`skills/light-kanban-worker/` is admitted into the first-party collection.
+All frozen acceptance criteria (AC-1…AC-14) are satisfied with appropriately
+labeled evidence: structural (contract, behavior, collection suites),
+installation (clean-copy discovery), behavioral (scenarios A–F against a real
+Light-Kanban server), invocation (fresh trigger/non-trigger probes), and
+review (Critic candidates F-001…F-004 disposed; repairs verified by the fresh
+Evaluator).
+
+## Completed work
+- Package: SKILL.md, agents/openai.yaml, references/api.md, contract +
+  behavior test suites with adversarial single-rule fixtures.
+- Collection synchronization: catalog, README, installation guide, maintenance
+  baseline, discovery/contract tests, bilingual guides, changelog.
+
+## Unfinished work (out of admission scope)
+- v0.1.4 release: tag publication, published-tag fresh-install verification
+  (`npx skills add LightDevCoder/skills#v0.1.4 …`), release receipt and
+  installation-verification records.
+- Light-Kanban repository: README/README_CN rewrite, Use Cases, v1.0.5
+  release preparation.
+
+## Risks
+- LAN cross-machine reachability is documented as a block rule, not
+  live-tested (Charter limitation).
+- The agent id "no guessing" rule is enforced by the executing agent reading
+  SKILL.md (instruction-level).
+
+## Reopen note
+Reopen with a recorded Change Proposal if the package capability, invocation
+policy, supported host, or required dependency changes; do not edit Charter
+revision 1 in place.

docs/skills/light-kanban-worker.md

@@ -79,12 +79,13 @@ outside the catalog. See the [admission evidence](../evidence/admissions/light-k
 
 ## Installation and discovery
 
-Install the published package with:
+Install the package with:
 
 ```text
 npx skills add LightDevCoder/skills --skill light-kanban-worker --yes --copy --agent '*'
 ```
 
-Refresh the agent host and confirm the Skill is discovered without relying on
-the source checkout. It is compatible with Light-Kanban v1.0.4+; v1.0.5 is the
-recommended integration version.
+(The pinned `#v0.1.4` form becomes the verified release command once the
+v0.1.4 release gates pass.) Refresh the agent host and confirm the Skill is
+discovered without relying on the source checkout. It is compatible with
+Light-Kanban v1.0.4+; v1.0.5 is the recommended integration version.