cc-devflow 4.5.13 → 4.5.14

package/.claude/skills/cc-act/SKILL.md

@@ -68,7 +68,7 @@ tool_budget:
 - `devflow/postmortems/INDEX.md`
 - `devflow/postmortems/incidents/<date>-<change-key>.md`
 
-其余状态和历史交给 Git。不要写额外过程文件或 JSON 文档。
+其余状态和历史交给 Git。不要写额外过程文件。
 
 ## Ship Modes
 

package/.claude/skills/cc-check/CHANGELOG.md

@@ -1,5 +1,11 @@
 # CC-Check Skill Changelog
 
+## v1.12.2 - 2026-05-14
+
+- restore the verification flow that was over-pruned during artifact minimization: reset contract, rerun reality, boundary checks, claim evidence, QA/test-quality review, diff/scope audit, and failure ownership
+- keep check output in the response and Git commit instead of reviving local verification process files
+- expand the short verdict shape with QA and diff sections so `pass` still requires fresh, mapped evidence
+
 ## v1.12.1 - 2026-05-13
 
 - compress the visible output contract to `Verdict / Evidence / Review / Route`

package/.claude/skills/cc-check/PLAYBOOK.md

@@ -11,7 +11,10 @@
 3. Re-run commands; do not inherit green claims from chat.
 4. Do not generate process files.
 5. Separate missing evidence from proven failure.
-6. Commit the completed Check stage when the environment finishes.
+6. Map every claim to evidence before verdict.
+7. Review test quality for behavior changes and bugfixes.
+8. Classify failure ownership before routing.
+9. Commit the completed Check stage when the environment finishes.
 
 ## Verdict Shape
 
@@ -26,6 +29,20 @@ Short response only:
 - Verdict: `pass` / `fail` / `blocked`
 - Evidence: commands, exit status, and proven claims
 - Review: clean, findings remain, or not reviewed
+- QA: feedback loop and test quality where applicable
+- Diff: scope match or drift
 - Route: `cc-act` / `cc-do` / `cc-investigate` / `cc-plan` / `stop`
 
 Do not write process files.
+
+## Reset Signals
+
+Stop and rebuild the verdict when:
+
+- output is inherited from chat
+- command output is stale
+- test is green but never proved Red
+- behavior is tested through private implementation only
+- failure owner is unknown
+- current diff does not match `task.md`
+- review finding is unresolved or stale

package/.claude/skills/cc-check/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: cc-check
-version: 1.12.1
+version: 1.12.2
 description: Use when a planned or investigated change needs fresh verification evidence and an honest pass/fail/blocked verdict before cc-act.
 triggers:
   - 验收这个需求
@@ -27,10 +27,15 @@ entry_gate:
   - Resolve the CLI with `../cc-dev/scripts/resolve-cc-devflow.sh require query workflow-context` when workflow query is needed; unsupported CLIs are blockers.
   - Read `task.md`, current Git diff, relevant code/tests, PR text when present, and fresh command output.
   - Re-run fresh commands instead of inheriting cc-do narration.
+  - Classify the current reality as pass-candidate, fail-candidate, or blocked-candidate before writing a verdict.
+  - Map every completion claim to command output, exit status, and key observation.
   - Do not create process files.
 exit_criteria:
   - Verdict is exactly pass, fail, or blocked.
   - Every passing statement cites fresh command output, exit status, and what claim it proves.
+  - Behavior changes and bugfixes include feedback-loop and test-quality review.
+  - Current diff is checked against `task.md` for missing scope, scope creep, and unintended file touch.
+  - Failures are classified as branch, baseline, environment, external, or unknown.
   - Missing evidence is separated from real failure.
   - After pass/fail/blocked, commit the stage state when repository policy allows it.
   - The next step is unambiguous: cc-act, cc-do, cc-investigate, cc-plan, or stop.
@@ -79,11 +84,63 @@ NO PASS WITHOUT FRESH EVIDENCE
    - `blocked`: required environment, auth, input, or evidence is missing.
 6. Commit the stage when this PDCA/IDCA environment completes.
 
+## Verification Phases
+
+1. Reset Contract：读取 `task.md`、当前 diff、相关代码 / 测试、PR 文本（如果存在），明确本轮验证对象。
+2. Re-run Reality：重新运行可信 gate，不继承 `cc-do` 叙述；记录命令、退出码、关键输出、skip / blocked 原因。
+3. Check Boundaries：检查 runtime、task completion、requirement diff、claim evidence、QA/test quality、review freshness、docs / UI / DX 影响。
+4. Freeze Verdict：只输出 `pass` / `fail` / `blocked`，并给出 route。
+
+任一阶段发现证据过期、边界矛盾、结论无法诚实成立，都先 reset 或 reroute，不硬凑 `pass`。
+
+## Evidence Layers
+
+`cc-check` 不是只看测试绿不绿。按风险选择必要层，至少说明未覆盖层的 skip reason：
+
+1. Runtime：test、lint、typecheck、build、脚本 gate。
+2. Task Completion：`task.md` 的任务是否真的完成，完成证据是否对应当前 diff。
+3. Requirement Diff：当前改动是否兑现需求，是否有 scope creep、missing requirement、意外文件触点。
+4. Claim Evidence：每个通过声明都有命令、退出码、观察和证明的 claim。
+5. QA Test Quality：red/green、public seam、mock boundary、fixture honesty、test-only API smell。
+6. Behavior Evidence：用户可见 expected / actual / reproduction steps 是否被当前反馈环覆盖。
+7. Review Freshness：本轮 review 是否覆盖当前 HEAD；未 review 要说清风险。
+8. Failure Ownership：失败归属 branch、baseline、environment、external 或 unknown。
+9. Docs / Browser / Operator：UI、CLI、docs、operator workflow 受影响时有证据或 skip reason。
+
+## Claim Evidence Matrix
+
+不要把所有绿色写成“测试过了”：
+
+| Claim | Required proof | Not enough |
+| --- | --- | --- |
+| Tests pass | current command, exit 0, failures 0 | old output |
+| Lint clean | current lint command, no errors | formatter only |
+| Build succeeds | build command exit 0 | tests only |
+| Bug fixed | original symptom or regression loop passes | code changed |
+| Regression test works | red -> green evidence | green only |
+| Requirements met | each task/acceptance mapped to proof | self-report |
+
+缺关键 claim 证据时，结论至少是 `blocked`。
+
+## Failure Ownership
+
+失败不能只写“红了”：
+
+- `branch`：本分支引入或当前改动相关。
+- `baseline`：有 base branch 或历史证据证明预先存在。
+- `environment`：缺依赖、权限、服务、密钥或平台条件。
+- `external`：外部系统或第三方服务失败。
+- `unknown`：归属不明；不能支撑 `pass`。
+
+每个失败写清 error name、证据、owner、rescue action。
+
 ## Output
 
 只输出短结论，不写过程文件：
 
 - Verdict: `pass` / `fail` / `blocked`
 - Evidence: command, exit status, and the claim it proves
-- Review: clean, findings remain, or not reviewed
+- Review: clean, findings remain, not reviewed, or skipped with reason
+- QA: feedback loop, test quality, and behavior evidence when applicable
+- Diff: scope match, missing scope, or scope drift
 - Route: `cc-act` / `cc-do` / `cc-investigate` / `cc-plan` / `stop`

package/.claude/skills/cc-check/references/gate-contract.md

@@ -16,6 +16,15 @@
 
 所有通过结论都必须来自本次新鲜证据；旧输出只能当线索，不能直接继承 verdict。
 
+## Verification Phases
+
+1. Reset Contract: read `task.md`, current diff, relevant code/tests, and PR text when present.
+2. Re-run Reality: run fresh commands and record exit code plus key observation.
+3. Check Boundaries: map task completion, requirement diff, claim evidence, QA/test quality, review freshness, and docs/UI/operator impact.
+4. Freeze Verdict: choose `pass`, `fail`, or `blocked` and route honestly.
+
+Skipping a relevant phase requires a concrete reason in the response.
+
 ## QA Feedback Loop
 
 行为变更和 bugfix 的 evidence 还必须说明反馈环：
@@ -27,6 +36,31 @@
 
 没有可信 loop 的 bugfix 默认不能 `pass`。
 
+## Claim Evidence
+
+Every claim needs proof:
+
+- tests pass: current command and exit 0
+- lint clean: current lint command and no errors
+- build succeeds: build command exit 0
+- bug fixed: original symptom or regression loop now passes
+- regression test works: red then green evidence
+- requirements met: each task or acceptance item mapped to proof
+
+Missing proof is `blocked`, not `pass`.
+
+## Failure Ownership
+
+Classify every failure:
+
+- `branch`
+- `baseline`
+- `environment`
+- `external`
+- `unknown`
+
+`unknown` cannot support `pass`. It needs a rescue action or a reroute.
+
 ## Reroute
 
 - `none`

package/.claude/skills/cc-check/references/review-contract.md

@@ -13,9 +13,20 @@ Use current evidence:
 - code and test inspection
 - cc-review findings from the current conversation when available
 
+## Review Layers
+
+- Task completion: every checked task in `task.md` has matching evidence.
+- Requirement diff: current diff matches approved scope and has no unexplained extra touch.
+- Critical pass: check data safety, injection, trust boundary, time/window safety, errors swallowed, and silent data loss when relevant.
+- Test quality: Red/Green proof, public seam, mock boundary, fixture honesty, and no test-only production API.
+- Docs/UI/operator: affected surfaces have evidence or explicit skip reason.
+- Freshness: review facts cover current HEAD, not an earlier diff.
+
 ## Gate Rules
 
 - Unfixed important findings route to `cc-do`.
 - Missing behavior evidence is `blocked`, not `pass`.
 - Stale command output is ignored.
 - Review facts belong in the response, PR brief, or Git commits, not in process files.
+- Low-confidence concerns are warnings unless evidence makes them blocking.
+- Scope or design invalidation routes to `cc-plan`; disproven root cause routes to `cc-investigate`.

package/.claude/skills/cc-do/CHANGELOG.md

@@ -1,5 +1,11 @@
 # CC-Do Skill Changelog
 
+## v1.7.2 - 2026-05-14
+
+- restore the execution discipline that was over-pruned during artifact minimization: tracer-bullet TDD, public seam Red evidence, mock/fixture honesty, Green minimality, refactor-after-green, and repeated-failure reroutes
+- keep execution proof in `task.md`, command output, final response, and Git commits instead of reviving per-task process files
+- strengthen recovery rules so stale chat memory and manual checkbox edits cannot substitute for task evidence
+
 ## v1.7.1 - 2026-05-13
 
 - simplify execution recovery so state comes from `task.md`, Git, and current repo truth

package/.claude/skills/cc-do/PLAYBOOK.md

@@ -9,12 +9,22 @@
 ## Task Gates
 
 1. Task selected from `task.md`
-2. Red observed or TDD exception recorded
-3. Green observed
-4. Refactor complete or unnecessary
-5. Verification run
-6. `mark-task-complete.sh` updates `task.md`
-7. Git commit created
+2. Task shape checked: one behavior, public seam, mock boundary, Green minimality
+3. Red observed for the expected behavior or TDD exception recorded
+4. Green observed with minimal production change
+5. Refactor complete or explicitly unnecessary
+6. Verification run
+7. `mark-task-complete.sh` updates `task.md`
+8. Git commit created
+
+## TDD Gates
+
+- No production code before an expected failing test unless the task records a TDD exception.
+- Red must fail because the target behavior is missing, not because of syntax, fixture, or mock mistakes.
+- Red must enter through a public seam; private implementation tests do not unlock Green.
+- Green must satisfy only the current Red.
+- Refactor happens only after Green and stays inside the current slice.
+- Fixture shortcuts and boundary mocks must not hide input contract problems.
 
 ## Recovery
 
@@ -28,6 +38,13 @@ Recover from Git and `task.md`:
 
 Do not recover from process files. Only `task.md`, Git, and current repo truth count.
 
+## Reroute Triggers
+
+- New evidence disproves root cause: `cc-investigate`
+- New evidence breaks scope, interface, or task contract: `cc-plan`
+- Repeated patch attempts fail after the same task: stop and reroute
+- Review feedback exceeds frozen scope: `cc-plan`
+
 ## Commit Rule
 
 Every completed task or execution environment gets its own commit. Split by behavior or layer when needed, but do not leave completed work uncommitted between stages.

package/.claude/skills/cc-do/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: cc-do
-version: 1.7.1
+version: 1.7.2
 description: Use when implementing frozen tasks, resuming interrupted work, applying an investigation handoff, or fixing review feedback inside the approved scope.
 triggers:
   - 开始做 T003
@@ -29,8 +29,11 @@ entry_gate:
   - Run `query workflow-context --change <changeId> --change-key <changeKey> --data-only --no-trace --compact`.
   - Read `task.md`, current Git status, and only the code/tests needed by the current task.
   - Reject execution if the task cannot be restated from `task.md` and repo evidence.
+  - Validate the task's execution shape before coding: Red test name, one observable behavior, public verification path, allowed boundary mocks, Green minimality guard, and refactor candidates.
 exit_criteria:
   - Current task has Red/Green evidence or a recorded TDD exception in `task.md`.
+  - Red evidence proves the target behavior through a public seam; Green evidence is minimal and does not pre-build future behavior.
+  - Refactor evidence names the smell removed or states why no refactor was needed.
   - Verification commands have been run or explicitly blocked.
   - Task status is updated through `scripts/mark-task-complete.sh`.
   - The completed task/environment is committed to Git.
@@ -58,7 +61,7 @@ tool_budget:
 - `devflow/changes/<change-key>/task.md` 中的任务状态
 - Git commit
 
-不要生成额外过程文件或 JSON 文档。失败和阻塞写在对用户的响应里；需要长期保留的失败教训交给 `cc-act` 写 incident postmortem。
+不要生成额外过程文件。失败和阻塞写在对用户的响应里；需要长期保留的失败教训交给 `cc-act` 写 incident postmortem。
 
 ## TDD Iron Law
 
@@ -79,6 +82,20 @@ NO PRODUCTION CODE WITHOUT A FAILING TEST FIRST
 7. Complete：运行 `scripts/mark-task-complete.sh --tasks devflow/changes/<change-key>/task.md --task <task-id>`。
 8. Commit：每个任务或执行环境完成后提交 Git commit。
 
+## Execution Discipline
+
+执行质量写进 `task.md` 的当前 task block、完成证据、最终回复和 Git commit，不写额外过程文件。
+
+1. 每次只推进一个 tracer bullet：一个用户可观察行为的 Red -> Green -> Refactor。
+2. Red 必须从公共接口、调用方流程、CLI/API/UI 路径或真实边界进入系统；私有函数、内部调用次数、临时结构断言不算 Red 证据。
+3. 一个 Red 只证明一个逻辑行为；bulk Red、实现步骤式测试名、直接查内部状态都要先修测试设计。
+4. Mock 只能放在系统边界：外部 API、时间、随机性、文件系统、网络或必要数据库边界；mock 自家模块默认说明 seam 有问题。
+5. Fixture 必须诚实：partial fixture、cast、generated stub、缺字段 payload 都要说明真实 contract 字段和测试填充字段。
+6. Green 只写当前红灯要求的最小实现；新增接口优先小而深，依赖从调用方传入，边界 adapter 用具体操作而不是 generic catch-all。
+7. Refactor 只能在 Green 后做，只清理当前 slice 暴露出的重复、命名、浅模块、长方法、feature envy、primitive obsession、错误处理或三层以上分支。
+8. 三次修补仍失败时，先质疑 `Root Cause Contract` 或 `Contract Summary`，reroute 到 `cc-investigate` / `cc-plan`，不要继续堆补丁。
+9. 完成脚本失败时，修缺失证据、依赖或 task block，不手改 checkbox 绕过。
+
 ## Parallel Rule
 
 只有任务显式允许并行、依赖已满足、触碰路径无重叠、且每个执行环境都能独立提交时，才并行。否则串行。

package/.claude/skills/cc-do/references/execution-recovery.md

@@ -15,6 +15,18 @@ Do not recover from process files. They are not part of the workflow.
 1. Read `task.md`.
 2. Run `git status --short`.
 3. Inspect recent commits.
-4. Select the first unchecked task.
-5. Re-run the task's verification command if previous output is stale.
-6. Continue or reroute.
+4. Select the first unchecked task whose dependencies are satisfied.
+5. Restate the task's goal, public seam, files, verification, and do-not-re-decide items.
+6. Re-run the task's verification command if previous output is stale.
+7. If Red/Green evidence is missing, rebuild the TDD loop before editing production code.
+8. Continue or reroute.
+
+## Bad Recovery Signals
+
+- Current task cannot be restated from `task.md`.
+- Dirty files do not match the selected task.
+- Prior green output is stale or came from chat memory.
+- Test seam is private or mock-heavy.
+- Completion script would only be satisfied by hand-editing status.
+
+Any of these forces a reset from Git, `task.md`, and fresh command output.

package/.claude/skills/cc-investigate/CHANGELOG.md

@@ -1,5 +1,11 @@
 # CC-Investigate Skill Changelog
 
+## v1.6.2 - 2026-05-14
+
+- restore the investigation flow that was over-pruned during artifact minimization: mode classification, feedback loop contract, evidence chain, hypothesis falsification, boundary/backward/reference evidence, diagnostic instrumentation, and correct test seam
+- keep root-cause proof and diagnose-only/reroute boundaries inside `task.md#Root Cause Contract` instead of reviving analysis, manifest, or report process files
+- update the task template and investigation contract so `cc-investigate` stays artifact-light without becoming a shallow bug checklist
+
 ## v1.6.1 - 2026-05-13
 
 - simplify the investigation artifact contract to `task.md` plus Git history

package/.claude/skills/cc-investigate/PLAYBOOK.md

@@ -15,18 +15,42 @@
 3. Write only `task.md`; Git records the stage.
 4. No process file beyond `task.md`.
 5. Reroute instead of mixing feature planning into bug investigation.
+6. Classify the investigation mode before tracing evidence.
+7. Record hypotheses with falsification methods and actual observations.
+8. Missing first bad state, original trigger, or counterfactual proof stops repair tasks.
 
 ## Investigation Modes
 
 - `reproduce-first`
+- `feedback-loop`
 - `diff-trace`
 - `boundary-probe`
 - `backward-trace`
 - `condition-wait`
 - `reference-compare`
 - `workflow-forensics`
+- `performance`
 - `diagnose-only`
 
+## Root Cause Contract Fields
+
+`task.md#Root Cause Contract` includes:
+
+- Investigation mode
+- Symptom, expected, actual, reproduction
+- Feedback loop contract
+- Evidence chain
+- Root Cause Proof Ladder
+- Hypothesis table
+- Boundary probe matrix when applicable
+- Backward trace chain when applicable
+- Reference comparison when applicable
+- Diagnostic instrumentation and cleanup when used
+- Repair boundary
+- Correct test seam
+- Prevention / escape reason
+- Diagnose-only or reroute next action when repair is not valid
+
 ## Exit
 
 `task.md` must tell `cc-do` what to fix, what not to touch, and what command proves the repair. Commit the Investigate stage before handing off.

package/.claude/skills/cc-investigate/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: cc-investigate
-version: 1.6.1
+version: 1.6.2
 description: Use when a bug, regression, broken task, or unexpected behavior needs root-cause investigation before coding resumes.
 triggers:
   - 帮我查这个 bug
@@ -29,10 +29,14 @@ entry_gate:
   - Assign a FIX change key through `next-change-key --prefix FIX --description "<short bug name>"`.
   - Enforce the Worktree Branch Contract before writing `task.md`.
   - Reproduce or build the closest honest feedback loop before naming root cause.
+  - Classify the investigation mode before tracing: reproduce-first, diff-trace, boundary-probe, backward-trace, reference-compare, condition-wait, workflow-forensics, performance, or diagnose-only.
   - Search relevant code, logs, recent history, and project postmortems before declaring the bug novel.
+  - Record hypotheses with falsification methods; do not collapse first intuition into root cause.
 exit_criteria:
   - "`task.md#Root Cause Contract` proves symptom site, first bad state, violated contract, original trigger, counterfactual proof, and escape reason."
+  - "`task.md#Root Cause Contract` records investigation mode, feedback loop, evidence chain, tested hypotheses, boundary/backward/reference evidence when applicable, correct test seam, and diagnose-only boundary when applicable."
   - "`task.md` contains the repair tasks needed by `cc-do`."
+  - "Evidence gaps produce Evidence Request, diagnose-only, or reroute tasks instead of fake repair tasks."
   - "No process file is created beyond `task.md`."
   - "Source roadmap progress is synced or explicitly skipped in the final response."
   - "Investigate-stage changes are committed to Git before handing off to `cc-do`."
@@ -57,7 +61,7 @@ tool_budget:
 
 - `devflow/changes/<change-key>/task.md`
 
-不要生成额外过程文件或 JSON 文档。Git commit 是阶段历史，`task.md` 是根因合同。
+不要生成额外过程文件。Git commit 是阶段历史，`task.md` 是根因合同。
 
 ## Iron Law
 
@@ -77,6 +81,36 @@ NO REPAIR WITHOUT A FROZEN ROOT-CAUSE CONTRACT
 6. Freeze：把根因、修复边界、测试 seam、allowed/forbidden files 写进 `task.md`。
 7. Commit：提交 Investigate 阶段，再交给 `cc-do`。
 
+## Investigation Modes
+
+先说清这是什么类问题，再决定证据路径：
+
+- `reproduce-first`：症状真实但复现不稳，先把现象钉死。
+- `feedback-loop`：已有复现但信号慢、宽或偶然，先让 loop 更快、更准、更可复跑。
+- `diff-trace`：明显 regression，先查最近变化和行为断点。
+- `boundary-probe`：多组件链路断裂，逐边界记录 input、output、config/env、state、verdict。
+- `backward-trace`：报错点很深或坏值来源不明，从 symptom site 追到 original trigger。
+- `reference-compare`：仓库里有相似可用路径，先列 working vs broken 差异。
+- `condition-wait`：flaky / sleep / timeout，找真实完成条件，不先加大等待。
+- `workflow-forensics`：坏在 artifact、Git、state、tool、permission 或流程。
+- `performance`：性能回归先拿 baseline、profile、query plan 或 bisect 信号。
+- `diagnose-only`：只做根因报告；task blocks 只能是证据交接、监控、人工动作或 reroute。
+
+## Evidence Flow
+
+调查结果写入 `task.md#Root Cause Contract`，不要拆新文件：
+
+1. Symptom Capture：用户看见什么、期望 / 实际、触发条件、影响面、复现入口。
+2. Feedback Loop Contract：loop type、命令或手动 driver、expected failing signal、actual failing signal、symptom match、runtime、determinism、failure rate、sharpening plan。
+3. Evidence Chain：按时间和数据流记录证据，不把代码阅读伪装成复现。
+4. Boundary Probe Matrix：多组件时记录每个边界的 input、output、config/env、state、verdict。
+5. Backward Trace Chain：deep error 时记录 immediate failure site、direct caller、caller chain、bad value origin、original trigger、rejected symptom-site fix。
+6. Reference Comparison：记录 similar working example、broken path、differences found、accepted hypotheses、ruled-out differences。
+7. Hypothesis Table：列 3-5 个候选，收敛到 1-3 个 active hypotheses；每个都有 support、counter-evidence、falsification method、expected observation、actual observation。
+8. Diagnostic Instrumentation：临时 probe 必须有 tag、location、question answered、command、expected signal、actual signal、cleanup requirement。
+9. Correct Test Seam：说明 regression test 是否覆盖真实触发链；只能测私有实现时，先记录设计缺口或 reroute。
+10. Repair Boundary：affected module、allowed files、forbidden files、blast radius、split-or-reroute decision。
+
 ## Root Cause Proof Ladder
 
 - L1 Symptom Site
@@ -88,6 +122,8 @@ NO REPAIR WITHOUT A FROZEN ROOT-CAUSE CONTRACT
 
 缺 L2、L4 或 L5 时，只能写 Evidence Request、diagnose-only 或 reroute。
 
+Escape reason 不能写成人工提醒；它必须变成 regression test、artifact guard、capability invariant、operator check 或明确 follow-up。
+
 ## Handoff
 
 退出时只说清：

package/.claude/skills/cc-investigate/assets/TASKS_TEMPLATE.md

@@ -13,6 +13,9 @@ Mode: investigation
 Profile: tiny | standard | deep
 Diagnosis:
 
+Investigation Mode:
+-
+
 Symptom:
 -
 
@@ -31,15 +34,79 @@ Root Cause:
 Evidence Chain:
 -
 
+Feedback Loop Contract:
+- Type:
+- Command / driver:
+- Expected failing signal:
+- Actual failing signal:
+- Symptom match:
+- Runtime / determinism / failure rate:
+- Sharpening plan:
+
+Root Cause Proof Ladder:
+| Level | Evidence | Verdict |
+|-------|----------|---------|
+| L1 Symptom Site |  |  |
+| L2 First Bad State |  |  |
+| L3 Violated Contract |  |  |
+| L4 Original Trigger |  |  |
+| L5 Counterfactual Proof |  |  |
+| L6 Escape Reason |  |  |
+
+Hypothesis Table:
+| Hypothesis | Support | Counter-evidence | Falsification method | Observation | Status |
+|------------|---------|------------------|----------------------|-------------|--------|
+| H1 |  |  |  |  | active |
+
+Boundary Probe Matrix:
+| Boundary | Input | Output | Config / env | State | Verdict |
+|----------|-------|--------|--------------|-------|---------|
+| N/A |  |  |  |  | not-applicable |
+
+Backward Trace Chain:
+- Immediate failure site:
+- Direct caller:
+- Caller chain:
+- Bad value origin:
+- Original trigger:
+- Why symptom-site fix is rejected:
+
+Reference Comparison:
+- Similar working example:
+- Broken path:
+- Differences found:
+- Accepted hypothesis:
+- Ruled-out differences:
+
+Diagnostic Instrumentation:
+- Probe tag:
+- Location:
+- Question answered:
+- Command:
+- Expected signal:
+- Actual signal:
+- Cleanup requirement:
+
 Repair Boundary:
 -
 
+Correct Test Seam:
+- Public seam:
+- Real trigger chain covered:
+- Mock boundary:
+- Private implementation assertions avoided:
+
 Verification:
 -
 
 Prevention:
 -
 
+Diagnose-Only / Reroute:
+- Applies: yes | no
+- Next action:
+- Why no repair task:
+
 Risk / Escalate If:
 -
 

package/.claude/skills/cc-investigate/references/investigation-contract.md

@@ -7,20 +7,52 @@
 
 ## Minimum Evidence
 
+- investigation mode
 - symptom
 - reproduction path
 - expected vs actual
+- feedback loop contract
 - code path
 - recent change signal
 - prior postmortem signal
+- evidence chain
+- tested hypotheses and falsification results
 - first bad state
 - violated contract
 - original trigger
 - counterfactual proof
 - escape reason
+- boundary probe, backward trace, or reference comparison when applicable
 - repair boundary
+- correct test seam
 - verification command
 
+## Investigation Modes
+
+- `reproduce-first`: stabilize the reported failure before naming root cause.
+- `feedback-loop`: sharpen a slow, broad, or flaky signal before proof.
+- `diff-trace`: inspect recent changes and behavioral breakpoints.
+- `boundary-probe`: inspect every component boundary before localizing repair.
+- `backward-trace`: trace from symptom site to bad value origin and original trigger.
+- `reference-compare`: compare working and broken paths before guessing.
+- `condition-wait`: replace sleeps/timeouts with real completion conditions.
+- `workflow-forensics`: classify artifact, Git, state, tool, permission, or process failure.
+- `performance`: collect baseline/profile/query/bisect evidence.
+- `diagnose-only`: produce evidence handoff, monitoring, human action, or reroute; do not invent repair tasks.
+
+## Root Cause Proof
+
+The contract must climb the ladder:
+
+1. Symptom Site
+2. First Bad State
+3. Violated Contract
+4. Original Trigger
+5. Counterfactual Proof
+6. Escape Reason
+
+Missing first bad state, original trigger, or counterfactual proof means no confirmed repair task. Return an Evidence Request, diagnose-only handoff, or reroute.
+
 ## Output Shape
 
 - `task.md#Root Cause Contract` is the human truth.

package/.claude/skills/cc-plan/CHANGELOG.md

@@ -1,5 +1,11 @@
 # CC-Plan Skill Changelog
 
+## v3.10.2 - 2026-05-14
+
+- restore the planning flow that was over-pruned during artifact minimization: requirement reality, system shape, interface/data contract, abstraction boundary, execution architecture, task contract, and final approval
+- keep decision questions, option comparison, PRD/user-story thinking, and engineering review inside `task.md#Contract Summary` instead of reviving separate process files
+- update the task template and planning contract so `cc-plan` remains artifact-light but still guides multi-round planning
+
 ## v3.10.1 - 2026-05-13
 
 - simplify the planning artifact contract to `task.md` plus Git history

package/.claude/skills/cc-plan/PLAYBOOK.md

@@ -14,7 +14,10 @@
 2. Git commits record Plan completion; do not create process files beyond `task.md`.
 3. Current branch must bind to the full change key before writing durable output.
 4. The task list must let `cc-do` continue without chat memory.
-5. Ask only blocking questions; otherwise choose from repo evidence.
+5. Ask only decisions that change scope, design, task split, interface, or verification; otherwise choose from repo evidence.
+6. Preserve planning thought inside `task.md#Contract Summary`: requirement reality, system shape, interface/data contract, abstraction boundary, execution architecture, task contract, and final approval.
+7. Non-trivial plans compare `minimal viable` and `ideal architecture`; tiny plans still record why the short path is enough.
+8. User-facing decisions use `D<N>` questions with recommendation, options, impact, and STOP.
 
 ## Required Task Fields
 
@@ -29,6 +32,20 @@ Each task block includes:
 - completion evidence
 - commit point if the task closes an execution environment
 
+## Contract Summary Fields
+
+`task.md#Contract Summary` includes:
+
+- Source handoff and repo evidence
+- Requirement reality
+- Decision questions and answers
+- Planning flow table
+- Minimal viable / ideal architecture comparison when non-trivial
+- Approved direction and non-goals
+- User stories and edge stories
+- Engineering review gate
+- Verification and escalation triggers
+
 ## Exit
 
 Run the smallest relevant validation for the plan artifacts, stage only owned changes, and commit the Plan stage. Then hand off to `cc-do`.