project-tiny-context-harness 0.2.57 → 0.2.59

package/assets/skills/plan_acceptance_checklist_compiler/SKILL.md

@@ -1,279 +1,307 @@
----
-name: plan_acceptance_checklist_compiler
-description: Use when the user provides or references a plan, Markdown file, RFC, implementation proposal, milestone plan, or long-task goal/target-mode task and explicitly asks for an acceptance checklist, completion definition, or goal/target-mode prompt. Triggers include acceptance checklist for this plan, goal-mode prompt for this implementation plan, target-mode prompt for this plan, 为这份方案生成验收清单, 整理这份方案输出目标模式文本. Do not trigger on standalone acceptance/目标模式/完成定义 terms without a plan-like source. This Skill writes temporary artifacts under tmp/ty-context/plan-acceptance and never executes the plan or marks completion.
----
-
-# Plan Acceptance Checklist Compiler
-
-## Package-Managed Boundary
-
-This Skill is generated by `ty-context sync` and owned by the Harness package. Do not edit the generated `plan_acceptance_checklist_compiler` Skill directly in a consumer project.
-
-Project-specific acceptance rules belong in `project_context/**`, `DESIGN.md` when visual design facts are durable, or a separate project-local Skill. Keep this package-managed Skill business-agnostic.
-
-## Purpose
-
-Convert a given long-task plan into a strict, project-context-aware acceptance checklist and a short goal/target-mode execution prompt.
-
-The output answers:
-
-```text
-For this exact plan, what must be true before a future executor can honestly say the task is complete?
-```
-
-This Skill is generic. Do not embed business-specific rules, provider-specific rules, product-domain assumptions, artifact schemas, API names, UI names, status names, or one-off project logic in this Skill. If the user's plan contains a specific domain, derive acceptance items from that plan and the project Context during the current invocation only.
-
-## Required Outputs
-
-Every completed invocation must produce:
-
-1. A preserved copy of the plan under `tmp/ty-context/plan-acceptance/`.
-2. A rigorous acceptance checklist derived from the plan and relevant project Context.
-3. A goal/target-mode prompt the user can paste directly into Codex.
-
-Exception: if the Context confirmation gate below triggers, stop after materializing the plan and reading enough Context to explain the uncertainty. Ask the user for confirmation before producing the checklist or goal/target-mode prompt.
-
-The goal/target-mode prompt must be no longer than 4000 characters. Use the user's language when practical. For Chinese requests, use this shape:
-
-```text
-实施计划: tmp/ty-context/plan-acceptance/<plan>.md
-可多开agent，agent名额不够了就关掉不用的。
-<验收清单>
-```
-
-For English requests, use this shape:
-
-```text
-Plan: tmp/ty-context/plan-acceptance/<plan>.md
-You may use multiple agents; if agent slots run low, close idle or unnecessary agents.
-<acceptance checklist>
-```
-
-This resource lifecycle line is part of the generated execution prompt. It authorizes parallel decomposition for the future executor while requiring unused agents to be closed when slots run low.
-
-## Non-Scope
-
-Do not execute the plan unless the user separately asks for execution.
-
-Do not modify product code, migrations, UI, API, tests, runtime artifacts, or durable Context while using this Skill.
-
-Do not run long validation, external probes, browser flows, cloud smoke tests, database writes, deployment commands, or other side-effect commands unless the user explicitly asks for validation execution.
-
-Do not mark any goal complete. This Skill defines acceptance; it does not complete acceptance.
-
-Do not lower the plan target to match current implementation.
-
-## Trigger Conditions
-
-Use this Skill when the user asks for any of the following:
-
-- Give a plan, Markdown file, RFC, milestone, or implementation proposal an acceptance checklist.
-- Combine project durable Context with a long-task plan to produce acceptance criteria.
-- Convert a long-range task plan into a checklist for Codex goal/target mode.
-- Generate a completion definition for "execute this plan until done."
-- Check whether a plan is missing acceptance conditions.
-- Combine user objective, plan steps, project Context, code entry points, and verification paths into an executable acceptance matrix.
-
-Typical trigger phrases:
-
-```text
-acceptance checklist for this plan
-goal-mode prompt for this implementation plan
-target-mode prompt for this plan
-completion definition for this plan
-long-task plan acceptance
-audit this plan for acceptance criteria
-整理这份方案，输出一份目标模式文本给我
-整理方案输出目标模式文本
-为这份方案生成验收清单
-方案验收清单
-长程任务方案验收
-实施计划验收清单
-结合项目 context 梳理方案验收
-为这份 md 生成目标模式验收文本
-把这份 md 整理成验收清单
-```
-
-Do not trigger from standalone broad phrases such as `goal mode`, `target mode`, `acceptance criteria`, `completion definition`, `目标模式文本`, `验收标准`, or `完成定义` unless the same user request also identifies a plan-like source.
-
-## Input Priority
-
-Build the checklist from these sources, in order:
-
-1. User's current instruction.
-2. The referenced or pasted plan.
-3. Relevant durable project Context under `project_context/**`.
-4. Repository guidance such as `AGENTS.md`, `README.md`, `DESIGN.md`, and relevant local Skills.
-5. Current code and tests, only to identify real surfaces, commands, routes, schemas, artifacts, entry points, and likely verification paths.
-6. Existing reports or artifacts, only to identify required evidence and invalidation risks. Existing artifacts are not proof unless the user explicitly asks to audit current completion.
-
-If plan and Context conflict, preserve the conflict in the checklist. Do not silently choose the easier side.
-
-## Step 1: Materialize The Plan Under `tmp/ty-context/plan-acceptance/`
-
-Before writing the checklist, write the user-specified plan into the repository `tmp/ty-context/plan-acceptance/` directory.
-
-Rules:
-
-- If `tmp/ty-context/plan-acceptance/` does not exist, create it.
-- If the user references a file outside `tmp/ty-context/plan-acceptance/`, copy its current content into `tmp/ty-context/plan-acceptance/<safe-plan-name>.md`.
-- If the user references a file already under `tmp/ty-context/plan-acceptance/`, use that path directly unless the user asks for a new copy.
-- If the user pasted the plan text, write the pasted plan exactly into `tmp/ty-context/plan-acceptance/<safe-plan-name>.md`.
-- Preserve the plan content. Do not summarize, normalize, reorder, translate, or edit it while materializing it.
-- Use a stable readable filename derived from the plan title, source filename, or user topic. Use lowercase letters, digits, hyphens, and `.md`.
-- If the source plan cannot be found or read, stop and report the missing source. Do not invent a plan.
-- The materialized plan is temporary execution input. It is not durable Context and not proof that any acceptance item passed.
-
+---
+name: plan_acceptance_checklist_compiler
+description: Use when the user provides or references a plan, Markdown file, RFC, implementation proposal, milestone plan, or long-task goal/target-mode task and explicitly asks for an acceptance checklist, completion definition, or goal/target-mode prompt. Triggers include acceptance checklist for this plan, goal-mode prompt for this implementation plan, target-mode prompt for this plan, 为这份方案生成验收清单, 整理这份方案输出目标模式文本. Do not trigger on standalone acceptance/目标模式/完成定义 terms without a plan-like source. This Skill writes temporary artifacts under tmp/ty-context/plan-acceptance and never executes the plan or marks completion.
+---
+
+# Plan Acceptance Checklist Compiler
+
+## Package-Managed Boundary
+
+This Skill is generated by `ty-context sync` and owned by the Harness package. Do not edit the generated `plan_acceptance_checklist_compiler` Skill directly in a consumer project.
+
+Project-specific acceptance rules belong in `project_context/**`, `DESIGN.md` when visual design facts are durable, or a separate project-local Skill. Keep this package-managed Skill business-agnostic.
+
+## Purpose
+
+Convert a given long-task plan into a strict, project-context-aware acceptance checklist and a short goal/target-mode execution prompt.
+
+The output answers:
+
+```text
+For this exact plan, what must be true before a future executor can honestly say the task is complete?
+```
+
+This Skill is generic. Do not embed business-specific rules, provider-specific rules, product-domain assumptions, artifact schemas, API names, UI names, status names, or one-off project logic in this Skill. If the user's plan contains a specific domain, derive acceptance items from that plan and the project Context during the current invocation only.
+
+## Required Outputs
+
+Every completed invocation must produce:
+
+1. A preserved copy of the plan under `tmp/ty-context/plan-acceptance/`.
+2. A rigorous acceptance checklist derived from the plan and relevant project Context.
+3. A goal/target-mode prompt the user can paste directly into Codex.
+
+Exception: if the Context confirmation gate below triggers, stop after materializing the plan and reading enough Context to explain the uncertainty. Ask the user for confirmation before producing the checklist or goal/target-mode prompt.
+
+The goal/target-mode prompt must be no longer than 3980 characters, including line breaks. This conservative budget keeps the prompt below Codex's 4000-character practical paste boundary even when character counting differs slightly. Keep the prompt information-dense: preserve required paths, core acceptance criteria, evidence rules, blockers and false-completion traps while using compact wording. Use the user's language when practical. For Chinese requests, use this shape:
+
+```text
+实施计划: tmp/ty-context/plan-acceptance/<plan>.md
+可多开agent，agent名额不够了就关掉不用的。
+<验收清单>
+```
+
+For English requests, use this shape:
+
+```text
+Plan: tmp/ty-context/plan-acceptance/<plan>.md
+You may use multiple agents; if agent slots run low, close idle or unnecessary agents.
+<acceptance checklist>
+```
+
+This resource lifecycle line is part of the generated execution prompt. It authorizes parallel decomposition for the future executor while requiring unused agents to be closed when slots run low.
+
+## Non-Scope
+
+Do not execute the plan unless the user separately asks for execution.
+
+Do not modify product code, migrations, UI, API, tests, runtime artifacts, or durable Context while using this Skill.
+
+Do not run long validation, external probes, browser flows, cloud smoke tests, database writes, deployment commands, or other side-effect commands unless the user explicitly asks for validation execution.
+
+Do not mark any goal complete. This Skill defines acceptance; it does not complete acceptance.
+
+Do not lower the plan target to match current implementation.
+
+## Trigger Conditions
+
+Use this Skill when the user asks for any of the following:
+
+- Give a plan, Markdown file, RFC, milestone, or implementation proposal an acceptance checklist.
+- Combine project durable Context with a long-task plan to produce acceptance criteria.
+- Convert a long-range task plan into a checklist for Codex goal/target mode.
+- Generate a completion definition for "execute this plan until done."
+- Check whether a plan is missing acceptance conditions.
+- Combine user objective, plan steps, project Context, code entry points, and verification paths into an executable acceptance matrix.
+
+Typical trigger phrases:
+
+```text
+acceptance checklist for this plan
+goal-mode prompt for this implementation plan
+target-mode prompt for this plan
+completion definition for this plan
+long-task plan acceptance
+audit this plan for acceptance criteria
+整理这份方案，输出一份目标模式文本给我
+整理方案输出目标模式文本
+为这份方案生成验收清单
+方案验收清单
+长程任务方案验收
+实施计划验收清单
+结合项目 context 梳理方案验收
+为这份 md 生成目标模式验收文本
+把这份 md 整理成验收清单
+```
+
+Do not trigger from standalone broad phrases such as `goal mode`, `target mode`, `acceptance criteria`, `completion definition`, `目标模式文本`, `验收标准`, or `完成定义` unless the same user request also identifies a plan-like source.
+
+## Input Priority
+
+Build the checklist from these sources, in order:
+
+1. User's current instruction.
+2. The referenced or pasted plan.
+3. Relevant durable project Context under `project_context/**`.
+4. Repository guidance such as `AGENTS.md`, `README.md`, `DESIGN.md`, and relevant local Skills.
+5. Current code and tests, only to identify real surfaces, commands, routes, schemas, artifacts, entry points, and likely verification paths.
+6. Existing reports or artifacts, only to identify required evidence and invalidation risks. Existing artifacts are not proof unless the user explicitly asks to audit current completion.
+
+If plan and Context conflict, preserve the conflict in the checklist. Do not silently choose the easier side.
+
+## Step 1: Materialize The Plan Under `tmp/ty-context/plan-acceptance/`
+
+Before writing the checklist, write the user-specified plan into the repository `tmp/ty-context/plan-acceptance/` directory.
+
+Rules:
+
+- If `tmp/ty-context/plan-acceptance/` does not exist, create it.
+- If the user references a file outside `tmp/ty-context/plan-acceptance/`, copy its current content into `tmp/ty-context/plan-acceptance/<safe-plan-name>.md`.
+- If the user references a file already under `tmp/ty-context/plan-acceptance/`, use that path directly unless the user asks for a new copy.
+- If the user pasted the plan text, write the pasted plan exactly into `tmp/ty-context/plan-acceptance/<safe-plan-name>.md`.
+- Preserve the plan content. Do not summarize, normalize, reorder, translate, or edit it while materializing it.
+- Use a stable readable filename derived from the plan title, source filename, or user topic. Use lowercase letters, digits, hyphens, and `.md`.
+- If the source plan cannot be found or read, stop and report the missing source. Do not invent a plan.
+- The materialized plan is temporary execution input. It is not durable Context and not proof that any acceptance item passed.
+
 Recommended paths:
 
 ```text
 tmp/ty-context/plan-acceptance/<plan-slug>.md
 tmp/ty-context/plan-acceptance/<plan-slug>-acceptance-checklist.md
+tmp/ty-context/plan-acceptance/<plan-slug>-local-audit.md
 ```
 
-## Step 2: Read Relevant Project Context
-
-Read only the Context needed for the plan's impacted surfaces. Use Context to identify what the project says the system should mean.
-
-Use code and tests to identify current implementation surfaces and verification commands. Code describes current state; it must not silently redefine the plan's completion definition.
-
-Extract:
-
-- Product or system contract.
-- Module boundaries.
-- API, data, artifact, runtime, UI, and test contracts.
-- Validation and deployment paths.
-- Security, privacy, redaction, and environment rules.
-- Known non-goals and forbidden dependencies.
-- Existing implementation entry points.
-
-## Step 3: Run The Context Confirmation Gate
-
-After reading the plan and relevant Context, decide whether generating an acceptance checklist would require guessing or silently accepting many durable fact changes. If yes, stop and ask the user to confirm details before continuing.
-
-Trigger this gate when any of these apply:
-
-- The plan appears to add or change many long-term facts, such as product ownership, module boundaries, API/schema semantics, data contracts, state semantics, verification/deployment entry points, UI information architecture, security/privacy rules, or cross-domain dependency rules.
-- The plan would likely require updates across multiple Context roles or areas, or more than a small number of durable Context files.
-- The plan conflicts with current Context and the conflict changes the completion definition rather than only implementation details.
-- The plan uses broad or strategic language whose meaning cannot be made falsifiable from the plan and Context alone, such as "rebuild the whole architecture", "unify every module", "production-grade", "fully integrated", "maximize capability", `重构整体架构`, `统一所有模块`, `生产级`, `长期方案`, `完全接入`, or similar.
-- The acceptance outcome depends on choosing between product/architecture alternatives that the user has not explicitly chosen.
-- External approvals, credentials, runtime environments, legal/compliance constraints, paid/provider access, or irreversible operational effects materially change what can be accepted.
-- You judge that proceeding would encode a durable assumption the user probably expects to review.
-
-When the gate triggers:
-
-- Do not compile the final checklist, write the checklist file, or generate the goal/target-mode prompt yet.
-- Ask 1-3 concise questions focused on the blocking durable assumptions.
-- Include the materialized plan path if already written.
-- State the specific Context files or Context roles that made confirmation necessary.
-- Offer a concrete default only when it is safe and reversible; otherwise ask for an explicit decision.
-- Resume from Step 2 after the user answers.
-
-Do not trigger this gate merely because the plan is large. If the Context impact is clear, bounded, and the checklist can represent the durable Context updates as acceptance items without guessing, continue.
-
-## Step 4: Compile The Acceptance Contract
-
-Before the detailed checklist, write a compact contract:
-
-```markdown
-## Acceptance Contract
-
-- Plan source: `tmp/ty-context/plan-acceptance/<plan>.md`
-- User objective: <objective>
-- Completion definition: <what must be true>
-- Core surfaces: <code/API/UI/data/runtime/artifact/test/context/etc.>
-- Explicit non-goals: <items>
-- External blockers: <accounts/session/credentials/network/cloud/manual approval/etc.>
-- Evidence standard: <accepted proof types>
-```
-
-If the plan uses broad words such as `all`, `complete`, `maximum`, `maximized`, `accepted`, `formal`, `production`, `real`, `current`, or `verified`, convert them into explicit inventory and evidence requirements.
-
-## Step 5: Build The Full Acceptance Checklist
-
-Treat the checklist as the set of falsifiable acceptance items a future executor must satisfy. It is not a progress log.
-
-Create a checklist table with these columns:
-
-```markdown
-| ID | Acceptance item | Source | Scope | Required evidence | Verification method | Fail / not accepted if |
-|---|---|---|---|---|---|---|
-```
-
-Column rules:
-
-- `ID`: stable ID such as `AC-01`; group by workstream if useful.
-- `Acceptance item`: one atomic condition that must be true.
-- `Source`: user instruction, plan section, Context file, code contract, schema, API, UI, test, or risk note.
-- `Scope`: one of `core`, `conditional`, `blocked-external`, `out-of-scope`, `nice-to-have`.
-- `Required evidence`: exact proof required.
-- `Verification method`: concrete command, check, inspection path, API call, UI check, artifact validation, or manual proof.
-- `Fail / not accepted if`: concrete invalidation case.
-
-Each item must be evidence-bound and falsifiable. Split combined requirements into separate rows.
-
-## Hard Blocker Handling
-
-Treat any unresolved required blocker as non-completion. A checklist may describe blocked acceptance work, but blocked work is still not accepted until the required evidence exists.
-
-Use `blocked-external` only when the item is required for full completion but depends on unavailable user input, credentials, account/session access, manual approval, external environment state, provider availability, legal/compliance decision, or another dependency the future executor cannot satisfy locally.
-
-For every `blocked-external` item, state:
-
-- The exact missing evidence.
-- Why the future executor cannot produce it locally.
-- The required user, owner, external system, or approval action.
-- The fallback or degraded path, if the plan or Context allows one.
-- The acceptance impact if the blocker remains unresolved.
-
-Do not mark blocked work as `out-of-scope`, `conditional`, or `nice-to-have` merely because it is hard or currently unavailable. Do not treat partial local work, preparation, mocks, old evidence, or an unexercised fallback as completion for a blocked required item.
-
+The local audit path is for the future goal/target-mode executor. This compiler may require the prompt to maintain it, but the file is a temporary recovery cache and not proof that any acceptance item passed.
+
+## Step 2: Read Relevant Project Context
+
+Read only the Context needed for the plan's impacted surfaces. Use Context to identify what the project says the system should mean.
+
+Use code and tests to identify current implementation surfaces and verification commands. Code describes current state; it must not silently redefine the plan's completion definition.
+
+Extract:
+
+- Product or system contract.
+- Module boundaries.
+- API, data, artifact, runtime, UI, and test contracts.
+- Validation and deployment paths.
+- Security, privacy, redaction, and environment rules.
+- Known non-goals and forbidden dependencies.
+- Existing implementation entry points.
+
+## Step 3: Run The Context Confirmation Gate
+
+After reading the plan and relevant Context, decide whether generating an acceptance checklist would require guessing or silently accepting many durable fact changes. If yes, stop and ask the user to confirm details before continuing.
+
+Trigger this gate when any of these apply:
+
+- The plan appears to add or change many long-term facts, such as product ownership, module boundaries, API/schema semantics, data contracts, state semantics, verification/deployment entry points, UI information architecture, security/privacy rules, or cross-domain dependency rules.
+- The plan would likely require updates across multiple Context roles or areas, or more than a small number of durable Context files.
+- The plan conflicts with current Context and the conflict changes the completion definition rather than only implementation details.
+- The plan uses broad or strategic language whose meaning cannot be made falsifiable from the plan and Context alone, such as "rebuild the whole architecture", "unify every module", "production-grade", "fully integrated", "maximize capability", `重构整体架构`, `统一所有模块`, `生产级`, `长期方案`, `完全接入`, or similar.
+- The acceptance outcome depends on choosing between product/architecture alternatives that the user has not explicitly chosen.
+- External approvals, credentials, runtime environments, legal/compliance constraints, paid/provider access, or irreversible operational effects materially change what can be accepted.
+- You judge that proceeding would encode a durable assumption the user probably expects to review.
+
+When the gate triggers:
+
+- Do not compile the final checklist, write the checklist file, or generate the goal/target-mode prompt yet.
+- Ask 1-3 concise questions focused on the blocking durable assumptions.
+- Include the materialized plan path if already written.
+- State the specific Context files or Context roles that made confirmation necessary.
+- Offer a concrete default only when it is safe and reversible; otherwise ask for an explicit decision.
+- Resume from Step 2 after the user answers.
+
+Do not trigger this gate merely because the plan is large. If the Context impact is clear, bounded, and the checklist can represent the durable Context updates as acceptance items without guessing, continue.
+
+## Step 4: Compile The Acceptance Contract
+
+Before the detailed checklist, write a compact contract:
+
+```markdown
+## Acceptance Contract
+
+- Plan source: `tmp/ty-context/plan-acceptance/<plan>.md`
+- User objective: <objective>
+- Completion definition: <what must be true>
+- Core surfaces: <code/API/UI/data/runtime/artifact/test/context/etc.>
+- Explicit non-goals: <items>
+- External blockers: <accounts/session/credentials/network/cloud/manual approval/etc.>
+- Evidence standard: <accepted proof types>
+```
+
+If the plan uses broad words such as `all`, `complete`, `maximum`, `maximized`, `accepted`, `formal`, `production`, `real`, `current`, or `verified`, convert them into explicit inventory and evidence requirements.
+
+## Step 5: Build The Full Acceptance Checklist
+
+Treat the checklist as the set of falsifiable acceptance items a future executor must satisfy. It is not a progress log.
+
+Create a checklist table with these columns:
+
+```markdown
+| ID | Acceptance item | Source | Scope | Required evidence | Verification method | Fail / not accepted if |
+|---|---|---|---|---|---|---|
+```
+
+Column rules:
+
+- `ID`: stable ID such as `AC-01`; group by workstream if useful.
+- `Acceptance item`: one atomic condition that must be true.
+- `Source`: user instruction, plan section, Context file, code contract, schema, API, UI, test, or risk note.
+- `Scope`: one of `core`, `conditional`, `blocked-external`, `out-of-scope`, `nice-to-have`.
+- `Required evidence`: exact proof required.
+- `Verification method`: concrete command, check, inspection path, API call, UI check, artifact validation, or manual proof.
+- `Fail / not accepted if`: concrete invalidation case.
+
+Each item must be evidence-bound and falsifiable. Split combined requirements into separate rows.
+
+## Hard Blocker Handling
+
+Treat any unresolved required blocker as non-completion. A checklist may describe blocked acceptance work, but blocked work is still not accepted until the required evidence exists.
+
+Use `blocked-external` only when the item is required for full completion but depends on unavailable user input, credentials, account/session access, manual approval, external environment state, provider availability, legal/compliance decision, or another dependency the future executor cannot satisfy locally.
+
+For every `blocked-external` item, state:
+
+- The exact missing evidence.
+- Why the future executor cannot produce it locally.
+- The required user, owner, external system, or approval action.
+- The fallback or degraded path, if the plan or Context allows one.
+- The acceptance impact if the blocker remains unresolved.
+
+Do not mark blocked work as `out-of-scope`, `conditional`, or `nice-to-have` merely because it is hard or currently unavailable. Do not treat partial local work, preparation, mocks, old evidence, or an unexercised fallback as completion for a blocked required item.
+
 The generated goal/target-mode prompt must tell the future executor:
 
 - If any executable `core` item remains, continue working it before asking the user to resolve blockers.
 - If the only remaining incomplete required items are hard blockers that cannot be satisfied locally, pause and wait for the user or external owner instead of marking the goal complete.
 - The handoff must list blocker cause, missing evidence, acceptance impact, and the exact user/external action needed to resume.
 
-## Generic Acceptance Dimensions
-
-Include only dimensions relevant to the plan. Do not mechanically add irrelevant categories.
-
-Consider these generic dimensions:
-
-- Objective and completion definition.
-- Scope inventory and coverage.
-- Domain / project Context conformance.
-- Code implementation behavior.
-- API or interface contract.
-- Data model, schema, migration, or persistence.
-- Runtime state, configuration, session, credential, environment, or degraded behavior.
-- Artifact generation, schema, freshness, provenance, and acceptance.
-- UI or user-visible projection.
-- Async job, worker, scheduler, queue, or background process.
-- Security, privacy, redaction, secrets, and access control.
-- Observability, logs, diagnostics, and operator visibility.
-- Performance, timeout, boundedness, pagination, and resource budget.
-- Backward compatibility and non-regression.
-- Failure states, blocked states, retries, recovery, and next actions.
-- Unit, integration, contract, smoke, browser, cloud, or production-like validation.
-- Documentation or durable Context update, only when the plan requires a durable fact change.
-
-## Evidence Layer Separation
-
+## Minimal User Blocker Protocol
+
+The generated goal/target-mode prompt must make user questions minimal and action-oriented when the plan says to do as much locally as possible.
+
+Before asking the user, the future executor must complete safe self-service discovery that is available in the repository or current environment:
+
+- Read relevant project Context, code, docs, tests, runtime examples, local artifacts and official documentation when current external rules matter.
+- Use local CLIs, APIs, Browser/Chrome/Computer Use or other tools only when appropriate, safe and allowed by the current tool and security policy.
+- Do not bypass user approval, account boundaries or sensitive-operation rules.
+
+When user input is still required, the blocker request must be the smallest concrete action list, not a research assignment. It must include:
+
+- What was already tried.
+- What is still missing.
+- The exact page, menu, path, field, button, command input or setting when it can be identified.
+- The minimum value or action the user needs to provide.
+- Sensitive material the user should not send, such as cookies, full pages, HAR files, password-store data, raw payloads, unrelated account data, screenshots when a value-only request is enough, or secrets in chat when a safer runtime-only setup path exists.
+- Acceptance impact if the blocker remains unresolved.
+- Allowed fallback, deferred or narrowed-scope option, if the plan or Context permits one.
+
+Do not replace this with broad requests such as "provide the token", "send the AppSecret" or "give me the configuration credentials" when a narrower page, field, runtime setup step or delegated user action would satisfy the blocker.
+
+## Generic Acceptance Dimensions
+
+Include only dimensions relevant to the plan. Do not mechanically add irrelevant categories.
+
+Consider these generic dimensions:
+
+- Objective and completion definition.
+- Scope inventory and coverage.
+- Domain / project Context conformance.
+- Code implementation behavior.
+- API or interface contract.
+- Data model, schema, migration, or persistence.
+- Runtime state, configuration, session, credential, environment, or degraded behavior.
+- Artifact generation, schema, freshness, provenance, and acceptance.
+- UI or user-visible projection.
+- Async job, worker, scheduler, queue, or background process.
+- Security, privacy, redaction, secrets, and access control.
+- Observability, logs, diagnostics, and operator visibility.
+- Performance, timeout, boundedness, pagination, and resource budget.
+- Backward compatibility and non-regression.
+- Failure states, blocked states, retries, recovery, and next actions.
+- Unit, integration, contract, smoke, browser, cloud, or production-like validation.
+- Documentation or durable Context update, only when the plan requires a durable fact change.
+
+## Evidence Layer Separation
+
 For evidence-heavy or validation-heavy plans, keep these layers separate:
 
 ```text
 infrastructure implemented
+runtime configured
 runtime exercised
-evidence collected
-evidence accepted
-system surfaces reflect evidence
-final validation passed
+artifact generated
+artifact accepted by validator
+API/UI reflects accepted evidence
+final gate/check command passed
 ```
 
 Do not merge these layers unless the plan explicitly treats them as one.
 
+For runtime, provider, artifact or external-evidence acceptance, distinguish adapter or infrastructure implementation, formal artifact production, scope alignment, redaction, replayability, validator acceptance, registry/API/UI projection, full gate/check success and any explicitly narrowed or deferred scope. A fallback path must be exercised before it can prove full completion.
+
 Examples of generic false completion:
 
 - Code exists, but runtime behavior was not exercised.
@@ -281,119 +309,129 @@ Examples of generic false completion:
 - Tests pass, but the required API/UI/runtime proof was not checked.
 - A partial sample ran, but the plan required full inventory coverage.
 - A document was updated, but the system state did not change.
-
-## Scope Labels
-
-Use these exact labels:
-
-- `core`: required to call the plan complete.
-- `conditional`: required only when a stated condition applies.
-- `blocked-external`: required for full completion but depends on unavailable external input.
-- `out-of-scope`: explicitly excluded by the user or plan.
-- `nice-to-have`: useful but not required; use sparingly.
-
-Do not mark an item out of scope only because it is difficult.
-Do not mark a `blocked-external` item complete until its required evidence exists.
-
-## Conflict Handling
-
-If the plan, Context, and code disagree, add a conflict table:
-
-```markdown
-| Conflict | Plan says | Context says | Code currently does | Acceptance impact | Required resolution |
-|---|---|---|---|---|---|
-```
-
-Do not resolve conflict by silently following current code.
-
-## False-Completion Traps
-
-After the checklist, include traps that a future executor must not mistake for completion.
-
-Use generic wording. Do not add business-specific traps to the Skill itself.
-
-Template:
-
-```markdown
-## False-Completion Traps
-
-- <activity> is not enough unless <acceptance proof> also exists.
-- <partial proof> cannot prove <full acceptance item>.
-- <old/stale/mismatched evidence> cannot prove current completion.
-- <surface A> passing does not prove <surface B> unless the checklist links them.
-```
-
-## Suggested Execution Order
-
-Suggest an execution order that prioritizes the highest-risk proof first:
-
-1. Confirm scope inventory and completion definition.
-2. Confirm schemas/contracts and invalid evidence rules.
-3. Implement or repair the core path.
-4. Produce or inspect required runtime/artifact/API/UI evidence.
-5. Run contract/integration/regression checks.
-6. Run build/typecheck/lint/smoke checks.
-7. Update temporary checklist status only if the user asked for status tracking.
-
-Do not put low-signal checks before the core acceptance proof when the plan is evidence-driven.
-
-## Checklist Self-Test
-
-Every generated checklist must pass this self-test:
-
-```markdown
-## Checklist Self-Test
-
-- [ ] Every original plan section is mapped to at least one acceptance item or explicitly marked non-goal.
-- [ ] Every broad word such as all / maximum / complete / formal / production / verified is converted into inventory + evidence.
-- [ ] Every core item has a verification method.
-- [ ] Every evidence item defines invalid evidence.
-- [ ] Code, API, UI, data, runtime, artifacts, and tests are separated when they prove different things.
-- [ ] External blockers are separated from local remaining work.
-- [ ] Hard blockers are treated as non-completion, with required user/external action and pause conditions.
-- [ ] Current implementation was not used to reduce the user's completion definition.
-- [ ] The checklist can be handed to another Codex session without hidden assumptions.
-```
-
-## Goal/Target-Mode Prompt Generation
-
-After compiling the checklist, produce a paste-ready goal/target-mode prompt.
-
-Hard requirements:
-
-- The prompt must be no longer than 4000 characters including line breaks.
-- The first line must identify the plan path.
-- Use `实施计划: <path>` for Chinese prompts and `Plan: <path>` for English prompts.
-- The second line must be a resource lifecycle instruction: `可多开agent，agent名额不够了就关掉不用的。` for Chinese prompts or `You may use multiple agents; if agent slots run low, close idle or unnecessary agents.` for English prompts.
-- The remaining content must be the acceptance checklist or a compact version of it.
+- A fallback exists in code, but the fallback was not configured or exercised.
+- Old, partial, smoke, dry-run or research evidence exists, but the current acceptance contract requires full current proof.
+
+## Scope Labels
+
+Use these exact labels:
+
+- `core`: required to call the plan complete.
+- `conditional`: required only when a stated condition applies.
+- `blocked-external`: required for full completion but depends on unavailable external input.
+- `out-of-scope`: explicitly excluded by the user or plan.
+- `nice-to-have`: useful but not required; use sparingly.
+
+Do not mark an item out of scope only because it is difficult.
+Do not mark a `blocked-external` item complete until its required evidence exists.
+
+## Conflict Handling
+
+If the plan, Context, and code disagree, add a conflict table:
+
+```markdown
+| Conflict | Plan says | Context says | Code currently does | Acceptance impact | Required resolution |
+|---|---|---|---|---|---|
+```
+
+Do not resolve conflict by silently following current code.
+
+## False-Completion Traps
+
+After the checklist, include traps that a future executor must not mistake for completion.
+
+Use generic wording. Do not add business-specific traps to the Skill itself.
+
+Template:
+
+```markdown
+## False-Completion Traps
+
+- <activity> is not enough unless <acceptance proof> also exists.
+- <partial proof> cannot prove <full acceptance item>.
+- <old/stale/mismatched evidence> cannot prove current completion.
+- <surface A> passing does not prove <surface B> unless the checklist links them.
+```
+
+## Suggested Execution Order
+
+Suggest an execution order that prioritizes the highest-risk proof first:
+
+1. Confirm scope inventory and completion definition.
+2. Confirm schemas/contracts and invalid evidence rules.
+3. Implement or repair the core path.
+4. Produce or inspect required runtime/artifact/API/UI evidence.
+5. Run contract/integration/regression checks.
+6. Run build/typecheck/lint/smoke checks.
+7. Update temporary checklist status only if the user asked for status tracking.
+
+Do not put low-signal checks before the core acceptance proof when the plan is evidence-driven.
+
+## Checklist Self-Test
+
+Every generated checklist must pass this self-test:
+
+```markdown
+## Checklist Self-Test
+
+- [ ] Every original plan section is mapped to at least one acceptance item or explicitly marked non-goal.
+- [ ] Every broad word such as all / maximum / complete / formal / production / verified is converted into inventory + evidence.
+- [ ] Every core item has a verification method.
+- [ ] Every evidence item defines invalid evidence.
+- [ ] Code, API, UI, data, runtime, artifacts, and tests are separated when they prove different things.
+- [ ] External blockers are separated from local remaining work.
+- [ ] Hard blockers are treated as non-completion, with required user/external action and pause conditions.
+- [ ] Current implementation was not used to reduce the user's completion definition.
+- [ ] The checklist can be handed to another Codex session without hidden assumptions.
+```
+
+## Goal/Target-Mode Prompt Generation
+
+After compiling the checklist, produce a paste-ready goal/target-mode prompt.
+
+Hard requirements:
+
+- The prompt must be no longer than 3980 characters including line breaks. Treat 3980 as the effective hard budget and preserve information density; do not drop required paths, core acceptance categories, blocker rules, evidence rules or false-completion traps merely to be short.
+- The first line must identify the plan path.
+- Use `实施计划: <path>` for Chinese prompts and `Plan: <path>` for English prompts.
+- The second line must be a resource lifecycle instruction: `可多开agent，agent名额不够了就关掉不用的。` for Chinese prompts or `You may use multiple agents; if agent slots run low, close idle or unnecessary agents.` for English prompts.
+- The remaining content must be the acceptance checklist or a compact version of it.
 - The prompt must be self-contained enough for goal/target-mode execution.
+- The prompt must identify a local audit path, normally `tmp/ty-context/plan-acceptance/<plan-slug>-local-audit.md`, and require the future executor to read it before resuming, keep it current during execution, and use it only as target-mode acceptance progress state.
+- The prompt must require the local audit to record overall status (`complete`, `incomplete`, `blocked` or `narrowed-scope-complete`), each core AC status and current evidence, commands with result/time/failure reason, artifact or evidence paths, blockers and missing evidence, acceptance impact, explicit deferred or narrowed scope, and stale/partial/smoke/dry-run/research evidence that cannot prove full completion.
+- The prompt must say that local audit is not Context, not product-quality proof, not a global task manager, and not a replacement for project tests, CI, review, human acceptance, Task Contract or workflow-contract `plan.md`.
+- The prompt must say that when a Task Contract or workflow-contract `plan.md` exists, each acceptance item execution still follows it and the repository's Tiny Context workflow contract.
 - Do not include explanatory preface inside the prompt.
 - Do not include Markdown tables inside the prompt if they make it too long.
 - Prefer short numbered items over verbose prose.
-- If the full checklist is too large, write the full checklist to `tmp/ty-context/plan-acceptance/<plan-slug>-acceptance-checklist.md`, then compress the goal/target-mode prompt while preserving all core acceptance categories.
+- If the full checklist is too large, write the full checklist to `tmp/ty-context/plan-acceptance/<plan-slug>-acceptance-checklist.md`, then compress the goal/target-mode prompt by increasing information density while preserving all core acceptance categories.
 - The compact prompt may reference the full checklist path, but it must still include the core completion criteria directly.
-
-Recommended compact Chinese prompt shape:
-
+
+Recommended compact Chinese prompt shape:
+
 ```text
 实施计划: tmp/ty-context/plan-acceptance/<plan-slug>.md
 可多开agent，agent名额不够了就关掉不用的。
+完整验收清单: tmp/ty-context/plan-acceptance/<plan-slug>-acceptance-checklist.md
+执行审计: tmp/ty-context/plan-acceptance/<plan-slug>-local-audit.md
 
 验收清单：
 AC1 <核心完成定义，包含验收证据>
-AC2 <范围/清单/覆盖要求>
-AC3 <Context/架构/边界要求>
-AC4 <核心实现行为要求>
-AC5 <数据/API/接口/契约要求>
-AC6 <运行态/配置/外部依赖/阻塞分类要求>
-AC7 <artifact/evidence/schema/freshness/provenance 要求>
-AC8 <UI/用户可见/API 投影一致性要求>
-AC9 <安全/隐私/脱敏/secret 要求>
+AC2 <范围/清单/覆盖要求>
+AC3 <Context/架构/边界要求>
+AC4 <核心实现行为要求>
+AC5 <数据/API/接口/契约要求>
+AC6 <运行态/配置/外部依赖/阻塞分类要求>
+AC7 <artifact/evidence/schema/freshness/provenance 要求>
+AC8 <UI/用户可见/API 投影一致性要求>
+AC9 <安全/隐私/脱敏/secret 要求>
 AC10 <测试/构建/集成/smoke/回归要求>
 AC11 <文档/Context 更新要求，仅在计划要求时执行>
-AC12 完成前审计：逐条对照实施计划；每个 core 项必须有当前证据；未跑验证必须明示；有可继续执行的 core 项不得标记完成；外部/强卡点必须写明原因、缺失证据、验收影响和下一步；若剩余未完成项只有无法本地解决的强卡点，暂停并等待用户/外部 owner，不能标记目标完成。
+AC12 维护执行审计：恢复执行先读 audit；记录总体状态、每个 AC 当前证据、命令/结果/时间、artifact/evidence 路径、blocker、deferred/narrowed scope、不能证明 full completion 的旧/部分/smoke/dry-run/research 证据；audit 不是 Context、完成证明、全局任务管理器，也不替代 Task Contract 或流程契约 plan.md。
+AC13 最小用户卡点：问用户前先完成安全自助发现；需要用户介入时只给最小动作清单，写明已尝试、缺失项、具体页面/菜单/字段/按钮、最小值/动作、不要发送的敏感信息、验收影响、fallback/deferred。
+AC14 完成前审计：逐条对照实施计划和完整 checklist；每个 core 项必须有当前证据；未跑验证必须明示；有可继续执行的 core 项不得标记完成；外部/强卡点必须写明原因、缺失证据、验收影响和下一步；若剩余未完成项只有无法本地解决的强卡点，暂停并等待用户/外部 owner，不能标记目标完成。
 
-禁止把以下内容当完成：只改代码、只更新计划、只跑部分测试、只生成旧/部分/不被当前契约接受的证据、只完成基础设施但未完成验收证据、强卡点未解除、API/UI/数据/测试之间仍矛盾。
+禁止把以下内容当完成：只改代码、只更新计划、只跑部分测试、只生成旧/部分/不被当前契约接受的证据、只完成基础设施但未完成验收证据、runtime 未配置/未演练、artifact 未被 validator 接受、API/UI 未反映验收证据、fallback 未演练、强卡点未解除、API/UI/数据/测试之间仍矛盾。
 ```
 
 Recommended compact English prompt shape:
@@ -401,51 +439,55 @@ Recommended compact English prompt shape:
 ```text
 Plan: tmp/ty-context/plan-acceptance/<plan-slug>.md
 You may use multiple agents; if agent slots run low, close idle or unnecessary agents.
+Full checklist: tmp/ty-context/plan-acceptance/<plan-slug>-acceptance-checklist.md
+Local audit: tmp/ty-context/plan-acceptance/<plan-slug>-local-audit.md
 
 Acceptance checklist:
 AC1 <core completion definition with required evidence>
-AC2 <scope inventory and coverage>
-AC3 <Context / architecture / boundary conformance>
-AC4 <core implementation behavior>
-AC5 <data / API / interface / contract requirements>
-AC6 <runtime / configuration / external dependency / blocker classification>
-AC7 <artifact / evidence / schema / freshness / provenance requirements>
-AC8 <UI / user-visible / API projection consistency>
+AC2 <scope inventory and coverage>
+AC3 <Context / architecture / boundary conformance>
+AC4 <core implementation behavior>
+AC5 <data / API / interface / contract requirements>
+AC6 <runtime / configuration / external dependency / blocker classification>
+AC7 <artifact / evidence / schema / freshness / provenance requirements>
+AC8 <UI / user-visible / API projection consistency>
 AC9 <security / privacy / redaction / secret handling>
 AC10 <test / build / integration / smoke / regression requirements>
 AC11 <documentation / Context updates only when required by the plan>
-AC12 Final audit: compare every item against the plan; every core item needs current evidence; missing validation must be stated; any executable core item left open means the task is not complete; external or hard blockers need cause, missing evidence, acceptance impact, and next action; if only locally unsatisfiable hard blockers remain, pause for the user or external owner instead of marking the goal complete.
+AC12 Maintain local audit: read it before resuming; record overall status, every AC's current evidence, commands/results/time, artifact/evidence paths, blockers, deferred/narrowed scope, and stale/partial/smoke/dry-run/research evidence that cannot prove full completion; audit is not Context, completion proof, a global task manager, or a replacement for Task Contract or workflow-contract plan.md.
+AC13 Minimal user blocker protocol: before asking the user, complete safe self-service discovery; when user action is needed, provide only the smallest action list with what was tried, missing item, exact page/menu/path/field/button, minimum value/action, sensitive material not to send, acceptance impact, and fallback/deferred option.
+AC14 Final audit: compare every item against the plan and full checklist; every core item needs current evidence; missing validation must be stated; any executable core item left open means the task is not complete; external or hard blockers need cause, missing evidence, acceptance impact, and next action; if only locally unsatisfiable hard blockers remain, pause for the user or external owner instead of marking the goal complete.
 
-Do not count these as completion: code-only changes, plan-only updates, partial tests, stale or partial evidence, infrastructure without acceptance proof, unresolved hard blockers, or contradictions between API/UI/data/tests.
+Do not count these as completion: code-only changes, plan-only updates, partial tests, stale or partial evidence, infrastructure without acceptance proof, runtime not configured/exercised, artifact not accepted by validator, API/UI not reflecting accepted evidence, unexercised fallback, unresolved hard blockers, or contradictions between API/UI/data/tests.
 ```
-
-Before final response, check the prompt length. If it exceeds 4000 characters, compress it and check again.
-
-## Final Response Requirements
-
-For completed checklist runs, the final response must include:
-
-- `tmp/ty-context/plan-acceptance/` plan path.
-- Full checklist path if a full checklist file was written.
-- The paste-ready goal/target-mode prompt in a code block.
-- Any unresolved ambiguity that affects checklist accuracy.
-
-If the Context confirmation gate triggers, the response must ask for the needed confirmation instead of including the checklist or goal/target-mode prompt.
-
-Do not claim that any acceptance item has passed unless the user explicitly asked for a current completion audit and current evidence was inspected.
-
-## Forbidden Behaviors
-
-Do not include concrete business-domain logic in this Skill.
-
-Do not hardcode project-specific provider names, API names, UI names, artifact paths, schemas, statuses, or commands in this Skill.
-
-Do not execute the generated goal/target-mode prompt.
-
-Do not use the checklist to mark the task complete.
-
-Do not hide missing source files, ambiguous plan scope, or external blockers.
-
-Do not rewrite the user's plan while copying it into `tmp/ty-context/plan-acceptance/`.
-
-Do not use a short sample, stale artifact, old result, or current implementation convenience to reduce a plan that asked for full completion.
+
+Before final response, check the prompt length. If it exceeds 3980 characters, compress by tightening wording and referring to the full checklist path while preserving required paths, core AC categories, blocker/evidence rules and false-completion traps; then check again.
+
+## Final Response Requirements
+
+For completed checklist runs, the final response must include:
+
+- `tmp/ty-context/plan-acceptance/` plan path.
+- Full checklist path if a full checklist file was written.
+- The paste-ready goal/target-mode prompt in a code block.
+- Any unresolved ambiguity that affects checklist accuracy.
+
+If the Context confirmation gate triggers, the response must ask for the needed confirmation instead of including the checklist or goal/target-mode prompt.
+
+Do not claim that any acceptance item has passed unless the user explicitly asked for a current completion audit and current evidence was inspected.
+
+## Forbidden Behaviors
+
+Do not include concrete business-domain logic in this Skill.
+
+Do not hardcode project-specific provider names, API names, UI names, artifact paths, schemas, statuses, or commands in this Skill.
+
+Do not execute the generated goal/target-mode prompt.
+
+Do not use the checklist to mark the task complete.
+
+Do not hide missing source files, ambiguous plan scope, or external blockers.
+
+Do not rewrite the user's plan while copying it into `tmp/ty-context/plan-acceptance/`.
+
+Do not use a short sample, stale artifact, old result, or current implementation convenience to reduce a plan that asked for full completion.