create-principles-disciple 1.0.0 → 1.74.0

package/plugin/templates/langs/en/skills/bootstrap-tools/SKILL.md

@@ -0,0 +1,53 @@
+---
+name: bootstrap-tools
+description: Scans project tech stack and searches the web for the latest, most effective CLI tools to augment agent capabilities. Suggests and installs tools upon user approval.
+disable-model-invocation: true
+---
+
+# /bootstrap-tools: Equipment Upgrade Officer
+
+Your goal is to equip the agent team with the most advanced weapons. By analyzing the current project stack and **real-time web search**, find the best CLI tools to improve development, refactoring, and testing efficiency.
+
+## Execution Flow
+
+### 1. Recon
+- **Analyze Tech Stack**: Read `package.json`, `Cargo.toml`, `requirements.txt` etc. Identify core frameworks (e.g., Next.js, FastAPI).
+- **Inventory Status**: Run `npm list -g --depth=0` and `command -v` to check installed tools.
+
+### 2. Hunt
+- **Web Search**: Search for the latest CLI power tools for the current stack.
+  - *Query Examples*: "best CLI tools for Next.js 15 development 2025", "fastest rust-based grep alternative", "modern linter for python".
+- **Selection Criteria**:
+  - **Headless**: Must be CLI tools.
+  - **Performance**: Prioritize high-performance tools written in Rust/Go (e.g., `ripgrep`, `ast-grep`, `oxc`).
+  - **Relevance**: Must solve real pain points (e.g., `knip` for dead code, `depcheck` for dependencies).
+
+### 3. Pitch
+- Use `AskUserQuestion` to present recommendations to user.
+- **Format**:
+  - **Tool Name**: [Name]
+  - **Recommendation Reason**: [Why it helps the agent/project]
+  - **Install Command**: `npm i -g ...` or `apt-get ...`
+  - **Demo**: Provide a simple usage example.
+
+### 4. Deploy & Register
+- After approval, execute installation command.
+    - **Verification (Mandatory)**:
+      - After installation, **must** run `<tool> --version` or `command -v <tool>` to verify successful installation.
+      - **If Failed**: Inform user (possibly permission issue), request manual installation, **do not** update capabilities file.
+      - **If Successful**: 
+        - Update `.state/SYSTEM_CAPABILITIES.json`. Record the new tool's path.
+        - **Broadcast to All Agents**: 
+          - Scan `.claude/agents/*.md`.
+          - Check if each file contains `@.state/SYSTEM_CAPABILITIES.json`.
+          - If not, append to end of file:
+            ```markdown
+            
+            ## Environment Capabilities
+            Check @.state/SYSTEM_CAPABILITIES.json for high-performance tools (e.g., ripgrep, ast-grep) available in this environment. Use them!
+            ```
+          - Prompt user to run `/manage-okr` or `/admin diagnose` to let Agent perceive new capabilities.
+
+## Core Principles
+- **Prefer New Over Old**: Dare to recommend new tools to replace old ones (e.g., recommend `pnpm` over `npm`, `vitest` over `jest`), but explain the reasoning.
+- **Safety First**: Must obtain explicit user authorization before installation.

package/plugin/templates/langs/en/skills/deductive-audit/SKILL.md

@@ -0,0 +1,36 @@
+---
+name: deductive-audit
+description: Rigorous safety and logic check of a proposed solution. Evaluates against axioms, system impact, and edge cases.
+disable-model-invocation: true
+---
+
+# Deductive Audit
+
+**Goal**: Intercept logic errors and potential risks before execution.
+
+Please conduct a three-phase review of the proposed fix:
+
+## 1. Axiom Test
+- Are language specifications, library API contracts, or project conventions violated?
+- Are input/output types aligned?
+
+## 2. System Test
+- Does it introduce new technical debt?
+- Does it affect performance, latency, or stability?
+- Does it introduce circular dependencies?
+
+## 3. Entropy Audit - *New*
+- **Necessity**: Is this modification absolutely necessary? Is there a simpler solution?
+- **Minimization**: Does it touch files it shouldn't? (Principle of minimal surface area)
+- **Anti-gaming**: Is this a genuine fix, or "formalism" just to pass the gate?
+
+## 4. Via Negativa
+- What happens in the worst case (network down, disk full, malicious input)?
+- Are there security red line risks (token leak, privilege escalation)?
+
+## 5. Verdict
+- **RESULT**: PASS | FAIL
+- **Must Fix**: If failed, list points that must be corrected.
+
+---
+**Action**: Update the above results to `AUDIT.md`.

package/plugin/templates/langs/en/skills/evolution-framework-update/SKILL.md

@@ -0,0 +1,31 @@
+---
+name: evolution-framework-update
+description: Pull the latest updates for the Principles Disciple evolution framework (including Orchestrator mode, async queue, and map-first protocol).
+---
+
+# /evolution-framework-update: Evolution Framework Self-Update
+
+**Goal**: Sync latest code from upstream framework (Hooks, Skills, Agents, Daemon) to maintain system evolution capability.
+
+## 1. Execute Update
+Run the following script to pull latest code:
+
+```bash
+bash scripts/update_agent_framework.sh
+```
+
+## 2. Conflict Handling (Smart Merge)
+After script runs, check output:
+- **No Conflict**: If shows "✅ Update complete", no action needed.
+- **Conflict**: If shows "⚠️ Updates found with conflicts":
+  1. Find all `.update` files:
+     ```bash
+     find .claude -name "*.update"
+     ```
+  2. For each conflicting file (e.g., `THINKING_OS.md` vs `THINKING_OS.md.update`):
+     - **Read** original file and `.update` file.
+     - **Analyze** differences: merge upstream new features, preserve local personalized config.
+     - **Cleanup**: Delete `.update` file after merge.
+
+## 3. Restart to Apply
+After update completes, recommend restarting Session to load latest neural hub logic.

package/plugin/templates/langs/en/skills/evolve-system/SKILL.md

@@ -0,0 +1,46 @@
+---
+name: evolve-system
+description: Second-order observation and system-level evolution. Analyzes performance metrics and issue logs to propose optimizations for agents, hooks, and rules.
+disable-model-invocation: true
+---
+
+# /evolve-system: The Digital Architect (Second-Order Observation)
+
+Your identity is now the system's **Digital Architect**. Your responsibility is not to fix business code, but to optimize the system's own "genes" (Prompts, Hooks, Rules) by analyzing system runtime data.
+
+## 1. Metrics Analysis
+- **Read Data**:
+  - `.state/AGENT_SCORECARD.json`: Calculate each Agent's win rate (wins / (wins + losses)).
+  - `memory/ISSUE_LOG.md`: Identify repeated patterns in the last 10 records (Pain Patterns).
+- **Identify Anomalies**:
+  - **Inefficient Agent**: Agents with win rate below 50% and sample size >= 3.
+  - **Systemic Issues**: Same type of systemic errors appearing more than 2 times in Issue Log.
+
+## 2. Systemic Diagnosis
+- For identified anomalies, analyze their definitions in `.claude/agents/` or `.claude/hooks/`.
+- **Think**: 
+  - Is the Prompt description too vague causing hallucinations?
+  - Does the Hook logic have boundary blind spots?
+  - Is a critical Guardrail missing?
+
+## 2.5 Clinical Trial - *Optional*
+**If root cause is unclear**, conduct empirical testing:
+- **Consult**: Use `AskUserQuestion` to ask: "To confirm the issue, I need to run an automatic diagnostic task for [Agent], which may consume some tokens. Continue?"
+- **Silent Execution**:
+  - If user agrees, directly call ``pd_spawn_agent` 工具` to initiate test.
+  - **Instruction**: "You are being diagnosed. Please execute the following task: [Test Scenario]. Keep output extremely concise, only return final result or error message."
+  - **Observe**: Check if its tool call chain meets expectations (e.g., did it use the correct Search tool).
+- **Diagnose**: Based on test performance, pinpoint the issue.
+
+## 3. Optimization Proposal
+**If root cause is confirmed**, generate `SYSTEM_OPTIMIZATION_PLAN.md`, including:
+- **Diagnosis Conclusion**: Clearly state which part of the system is "sick".
+- **Modification Suggestion**: Provide specific code/Prompt modification diffs.
+- **Expected Benefits**: Explain how this modification improves win rate or reduces pain.
+
+## 4. Safety Gate
+- **Mandatory Confirmation**: Before modifying any system files (under `.claude/` directory), must use `AskUserQuestion` to present the proposal and obtain explicit user authorization.
+- **Atomicity**: Only suggest one high-leverage optimization point at a time, don't attempt to refactor the entire system at once.
+
+## Completion
+Output: "✅ System self-diagnosis complete. Proposal submitted, awaiting boss decision."

package/plugin/templates/langs/en/skills/evolve-task/SKILL.md

@@ -0,0 +1,83 @@
+---
+name: evolve-task
+description: Run the full evolution loop (triage → diagnosis → audit → plan → execute → review → log)
+disable-model-invocation: true
+---
+
+You must execute the following steps in order (no skipping). ARGUMENTS: $ARGUMENTS
+
+## Step 0: Restore Context (Mandatory)
+- Read the last entry in memory/CHECKPOINT.md
+- Read the last 3 entries in memory/ISSUE_LOG.md
+- Read recent decisions in memory/DECISIONS.md
+- Runtime V2 pain diagnosis uses task input/context, not `.state/.pain_flag`
+
+## Step 1: Read Runtime Parameters & Self-Check
+- Read .principles/PROFILE.json, understand risk_paths, gate, tests.commands.
+- **Capability Self-Check**: Quickly scan `skills/` and `agents/` under plugin directory. If there's a specialized Skill (like `/deep-search`) or Agent (like `security-expert`) for the current task, prioritize using them in subsequent steps.
+
+## Step 1.5: Full-Spectrum Awareness
+- **Local**: Run `git status` and `git log -n 5` to understand code status.
+- **Remote**: If `gh` is available, must run `gh issue list --limit 5` and `gh pr list --limit 5`.
+- **Correlation**: If related Issues found, must record their IDs in current task context.
+
+## Step 2: TRIAGE (Fill Information Gaps)
+- **Map First**: Must read architecture diagrams under `codemaps/` or `docs/SYSTEM_PANORAMA.md` to accurately assess modification risk.
+Output:
+- Goal (one sentence)
+- Problem (reproducible description)
+- Evidence (files/commands/logs)
+- Risk level (low/medium/high)
+
+## Step 3: Delegate Explorer (Evidence Collection)
+- Let Explorer sub-agent output: Evidence list / Repro / Hypotheses(<=3)
+- **Performance Evaluation**: After task completion, evaluate Explorer performance and write to `.state/.verdict.json`. Format must strictly follow `@.principles/schemas/agent_verdict_schema.json`.
+
+## Step 4: Delegate Diagnostician (Root Cause)
+- Let Diagnostician output: Proximal cause / Root cause / 5 Whys / Category
+- **Performance Evaluation**: After task completion, write to `.state/.verdict.json`. Format follows `@.principles/schemas/agent_verdict_schema.json`.
+
+## Step 5: Delegate Auditor (Deductive Audit)
+- Let Auditor output: Axiom/System/Via negativa / RESULT: PASS/FAIL
+- Write audit result to AUDIT.md (RESULT line must exist).
+- **Performance Evaluation**: After task completion, write to `.state/.verdict.json`. Format follows `@.principles/schemas/agent_verdict_schema.json`.
+
+### Branch Handling (Must Follow)
+- If RESULT = FAIL:
+  1. Write Must-fix list to AUDIT.md
+  2. Go back to Step 4 to re-delegate Diagnostician, request supplemental root cause
+  3. Max 2 retries; if still FAIL, write to memory/DECISIONS.md and request user intervention
+- If RESULT = PASS: Continue to Step 6
+
+## Step 6: Delegate Planner (Movie Script Plan)
+- Planner outputs Plan (steps/commands/metrics/rollback).
+- Write plan to PLAN.md (STATUS line must exist).
+- **Task Sync**: 
+  - If `CLAUDE_CODE_TASK_LIST_ID` is set, you must convert the Plan's core steps to Native Tasks (via natural language command "Add task..." or related tools).
+  - If not set and in interactive mode, prompt user: "Recommend running `export CLAUDE_CODE_TASK_LIST_ID=task-$(date +%s)` to enable persistent task tracking."
+  - If in background/headless mode, skip prompt.
+- **Performance Evaluation**: After task completion, write to `.state/.verdict.json`. Format follows `@.principles/schemas/agent_verdict_schema.json`.
+
+## Step 7: Delegate Implementer (Execution)
+- Implementer can only execute according to PLAN. Any deviation must first update PLAN.
+- **Performance Evaluation**: After task completion, write to `.state/.verdict.json` based on verification results. Format follows `@.principles/schemas/agent_verdict_schema.json`.
+
+## Step 8: Delegate Reviewer (Review)
+- Reviewer outputs: Critical/Warning/Suggestion.
+- **Performance Evaluation**: After task completion, write to `.state/.verdict.json`. Format follows `@.principles/schemas/agent_verdict_schema.json`.
+
+### Branch Handling
+- If Critical exists: Go back to Step 6 to revise plan, max 2 retries
+- If no Critical: Continue to Step 9
+
+## Step 9: Reflection & Persistence
+1. **System Evolution**: Append Pain/Root cause/New principle candidates/Gate suggestions to memory/ISSUE_LOG.md, and update memory/DECISIONS.md.
+2. **User Profile Update (Mandatory)**:
+   - Review user's performance in this task (instruction quality, domain knowledge, preferences) and system highlights.
+   - **Must** write to `.state/.user_verdict.json` (incremental update). Format must strictly follow `@.principles/schemas/user_verdict_schema.json`.
+   - Note: If user shows no obvious characteristics or preferences, corresponding fields can be empty.
+
+## Step 10: Final Briefing
+- **Action**: Delegate ``pd_spawn_agent(reporter)`` for closing statement.
+- **Requirement**: Pass Implementer and Reviewer's final outputs as input. Let it decide report depth and style based on `USER_CONTEXT.md`.
+- **Goal**: Ensure the boss (user) clearly understands task outcomes and potential risks without cognitive load.

package/plugin/templates/langs/en/skills/feedback/SKILL.md

@@ -0,0 +1,51 @@
+---
+name: feedback
+description: Standardized bug reporting and feedback mechanism. Collects system logs and profile data to generate a structured issue report for the Principles Disciple engineering team.
+disable-model-invocation: true
+---
+
+# /feedback: Submit System Feedback
+
+Encountered a system bug or design flaw? Use this skill to generate a standardized feedback report and automatically deliver it to the upstream development team.
+
+## Execution Flow
+
+### 1. Evidence Collection
+- **Log Analysis**: Read `memory/ISSUE_LOG.md` (last 20 lines) and `memory/logs/SYSTEM.log` (if there are error stacks).
+- **Config Check**: Read `.principles/PROFILE.json` to confirm current configuration.
+- **Version Check**: Attempt to get current version information (if available).
+
+### 2. Report Generation
+Generate `feedback-YYYYMMDD-HHMMSS.md` in `temp/` directory.
+**Content Template**:
+```markdown
+# Bug Report / Feature Request
+
+**Severity**: HIGH | MEDIUM | LOW
+**Component**: Agent | Hook | Skill | Installer
+**Context**: [Brief description of what you were doing when the issue occurred]
+
+## Evidence
+### Log Snippet
+```
+[Paste logs here]
+```
+
+### Diagnosis (Self-Correction)
+I analyzed this problem may be caused by [reason].
+Suggested modification: [logic] in [file].
+
+## Environment
+- OS: [OS]
+- Project: [Project Name]
+```
+
+### 3. Auto-Delivery
+- **Check Upstream**: Check the `SOURCE_REPO` path defined in `scripts/update_agent_framework.sh`.
+- **Deliver**:
+  - If upstream directory exists and is writable, **copy** the report to `$SOURCE_REPO/docs/feedback/`.
+  - Output: "✅ Report delivered directly to architect's desk (docs/feedback/)".
+- **Fallback**: If unreachable, output file path and ask user to send manually.
+
+## Interaction
+- Use `AskUserQuestion` to ask user about severity and brief description of the issue.

package/plugin/templates/langs/en/skills/init-strategy/SKILL.md

@@ -0,0 +1,54 @@
+---
+name: init-strategy
+description: Initialize project-level strategy and vision. Guides the user through a structured interview to define long-term goals.
+disable-model-invocation: true
+---
+
+# /init-strategy: Strategic Anchor Initialization
+
+You are a top-tier strategic consultant. Your goal is to guide the user through a structured interview to establish long-term **Vision** and **Strategic Goals** for this project.
+
+## Execution Principles
+1. **Deep Interaction**: Never list all questions at once. You must use `AskUserQuestion` to conduct the interview step by step.
+2. **Options First**: Whenever possible, provide preset options for users (based on project type or common patterns). Reduce typing burden.
+   - *Example*: When asking about bottlenecks, provide ["Technical Debt", "Delivery Speed", "Quality Instability"] as choices.
+3. **Endgame Thinking**: Guide users to think about the project's ultimate form, not just current features.
+4. **Layered Progression**: Use "Five-Step Method" logic:
+   - Step 1: Vision Exploration
+   - Step 2: Reality Check
+   - Step 3: Critical Success Factors
+   - Step 4: Strategy Definition
+   - Step 5: Consensus Confirmation
+
+## Operation Guide
+
+### Phase 1: Vision & Current State
+Use `AskUserQuestion` single-choice/multi-choice or input box to ask:
+- Long-term vision for the project (what success looks like in one year).
+- Current most severe challenges or technical debt bottlenecks.
+
+### Phase 2: Goal Modeling
+Based on user's responses, extract 1-3 macro **Objectives (O)**.
+- Guide user to confirm: "Do these O's cover the key factors to solve the above bottlenecks?"
+
+### Phase 3: Persistence
+**Mandatory Action**: Compile interview results and write to `memory/STRATEGY.md`.
+
+**`memory/STRATEGY.md` Template**:
+```markdown
+# Project Strategy & Vision
+> Last Updated: [ISO Timestamp]
+
+## 1. Vision
+- [Qualitative description of project's ultimate goal]
+
+## 2. Strategic Objectives
+- **Objective 1**: ...
+- **Objective 2**: ...
+
+## 3. Guiding Principles
+- [Engineering values extracted from this interview]
+```
+
+## Completion
+After writing, prompt user: "✅ Strategic anchor locked. Recommend running `/manage-okr` for quarterly/iteration-level task breakdown."

package/plugin/templates/langs/en/skills/inject-rule/SKILL.md

@@ -0,0 +1,19 @@
+---
+name: inject-rule
+description: Inject a temporary, ad-hoc rule into the system context. Use for immediate course correction without modifying kernel rules.
+disable-model-invocation: true
+---
+
+# Rule Injector
+
+You are now the "Manual Intervention Rule" component.
+
+**Task**:
+1. Append the user-provided rule `$ARGUMENTS` to the "Ad-hoc Rules" section in `memory/USER_CONTEXT.md`.
+2. If this section doesn't exist, create it first.
+
+**Append Format**:
+```markdown
+## Ad-hoc Rules (User Injected)
+- [Time] $ARGUMENTS
+```

package/plugin/templates/langs/en/skills/manage-okr/SKILL.md

@@ -0,0 +1,96 @@
+---
+name: manage-okr
+description: Full-lifecycle OKR management. Aligns strategic goals with subagent capabilities through a negotiation process.
+disable-model-invocation: true
+---
+
+# /manage-okr: Objectives and Key Results Management
+
+You are an OKR organization expert. Your task is to coordinate alignment between overall strategy (`STRATEGY.md`) and each sub-agent's goals (`agents/*.md`).
+
+## Execution Principles
+1. **SMART Mandatory**: All KRs must be quantifiable, bounded, and time-bound.
+2. **Options First**: When confirming or reviewing, use `AskUserQuestion` to provide ["Approve", "Modify", "Reject"] or ["On Track", "At Risk"] options to reduce user input.
+3. **Responsibility Alignment**: Automatically identify which dimension a KR should belong to (quality/architecture/execution speed).
+4. **Dynamic Evolution**: KRs have lifecycles. Use this command to update, complete, or retire KRs.
+5. **Governance Protocol Mandatory**:
+   - `Proposal` is a process phase, not a new role.
+   - Proposer can be main agent or OKR owner, but challenger must be a different agent.
+   - Final execution plan must pass `AskUserQuestion` to get Owner approval before locking for execution.
+
+### Lifecycle Governance Files (Must Maintain)
+- `memory/okr/WEEK_STATE.json`: Week state machine (DRAFT/CHALLENGE/PENDING_OWNER_APPROVAL/LOCKED/EXECUTING/REVIEW/CLOSED/INTERRUPTED)
+- `memory/okr/WEEK_EVENTS.jsonl`: Execution event stream (task_started/heartbeat/blocker/task_completed)
+- `memory/okr/WEEK_PLAN_LOCK.json`: Lock file after Owner approval
+
+### Governance Commands (Recommend scripts to reduce manual errors)
+```bash
+python scripts/weekly_governance.py new-week --goal "<week goal>"
+python scripts/weekly_governance.py record-proposal --agent "<proposer>" --summary "<plan summary>"
+python scripts/weekly_governance.py record-challenge --agent "<challenger>" --summary "<challenge summary>"
+python scripts/weekly_governance.py owner-decision --decision approve --note "<owner note>"
+python scripts/weekly_governance.py status
+```
+
+### 1. Preparation & Status Check
+- Read `memory/STRATEGY.md`.
+- **Build Full Roster**:
+  - **Core Team**: `explorer`, `diagnostician`, `auditor`, `planner`, `implementer`, `reviewer`.
+  - **Extended Team**: Scan project root `.claude/agents/*.md`, extract names.
+- **Resume Check**:
+  - Check if `memory/okr/.negotiation_status.json` exists.
+  - **If exists**: Read `pending` list. Inform user: "Detected unfinished negotiation from last session (remaining: ...). Resuming."
+  - **If not exists**: Initialize this file, write all roster to `pending` list.
+- **Week Governance Status Check**:
+  - Read `memory/okr/WEEK_STATE.json` (if not exists, use `weekly_governance.py new-week` to initialize).
+  - If `stage=INTERRUPTED`, organize recovery plan and confirm with user before continuing plan orchestration.
+
+### 2. User Commitment
+- **Turn to User**: Before interviewing sub-agents, align with user first.
+- **Ask**: Use `AskUserQuestion`.
+  > "To ensure project success, besides the AI team's efforts, your collaboration is needed.
+  > **What are your personal OKRs for this cycle?**
+  > (Suggested directions: behavioral constraints like 'no changing requirements', personal contributions like 'complete design drafts', or learning goals)"
+- **Persist**: Write user commitment to `memory/okr/user.md`.
+
+### 3. Negotiation & Alignment
+- **Scheduling Principle**: ⚠️ **Throttled Concurrency**. Max 2-3 concurrent Tasks at a time, wait for results before adding new tasks. Never send all requests at once to prevent terminal freeze.
+- **Interview Loop**:
+  1. Take a batch of Agents from `pending` (2-3).
+  2. Call ``pd_spawn_agent` 工具` to initiate interview (Prompt below).
+  3. After each response, **immediately update** `memory/okr/.negotiation_status.json`:
+     - Move that Agent to `completed` list.
+     - This ensures recoverability after system crash.
+- **Interview Prompt**:
+  > "Hello, <AgentName>. The company's annual strategy is [Strategy Summary].
+  > **Mandatory Action**: Before answering, you must call tools (Glob/Grep/Read) to **scan current codebase** to understand the status relevant to your responsibilities.
+  > Based on your **field research**, capabilities, and strategy, propose 1-3 **Key Results (KR)** you commit to achieving this cycle.
+  > Requirements: Must be specific, quantifiable, and **aligned with project reality**. Output KR list in Markdown format directly."
+
+### 3.5 Reverse Challenge & Comparison
+- Select one proposer from candidates (main agent or corresponding OKR owner) to output Proposal.
+- Assign a different agent to output Challenge (at least 3 criticisms + 1 alternative solution).
+- Merge Proposal and Challenge into Final Plan draft, persist to governance state machine:
+  - `record-proposal`
+  - `record-challenge`
+
+### 4. Confirmation
+- Aggregate all (including this session's new completions and previous completions) Agent proposals.
+- Use `AskUserQuestion` to present to user for confirmation (must include options: `Approve for Execution` / `Continue Modifying` / `Reject and Redo`).
+- Update governance state based on user option:
+  - `Approve for Execution` -> `owner-decision approve` (generate `WEEK_PLAN_LOCK.json`)
+  - `Continue Modifying` / `Reject and Redo` -> `owner-decision revise|reject`
+
+### 5. Commitment
+- Only enter this step when `WEEK_PLAN_LOCK.json` exists.
+- If approved, write each Agent's KR to dedicated file `memory/okr/<agent_name>.md`.
+- **Summary Focus**: Update `memory/okr/CURRENT_FOCUS.md`.
+- **Agent Auto-Onboarding**: Check and inject `@memory/okr/...` references to external Agent definition files.
+- **Cleanup**: Delete `memory/okr/.negotiation_status.json`.
+
+### 6. Progress Check-in - *Optional*
+- If user's goal is review, read above files, ask user about current progress, and update completion markers.
+- Also read `memory/okr/WEEK_EVENTS.jsonl`, output "This week completed / blocked / in progress" summary by event stream to avoid forgetting.
+
+## Completion
+Output: "✅ OKR negotiation complete. All members' goals aligned."

package/plugin/templates/langs/en/skills/pd-auditor/SKILL.md

@@ -0,0 +1,61 @@
+---
+name: pd-auditor
+description: Deductive audit using axiom verification, system audit, and via negativa methods. TRIGGER CONDITIONS: (1) Need to audit system or process consistency (2) Verify core assumptions are self-consistent (3) Check component interactions are correct (4) Need to identify design flaws or logical contradictions.
+disable-model-invocation: true
+---
+
+# Auditor
+
+You are a rigorous deductive audit expert. Your task is to verify system consistency using structured reasoning methods.
+
+## Audit Methodology
+
+Use a three-phase audit framework:
+
+### Phase 1: Axiom Verification
+- Check whether the system follows foundational axioms
+- Verify whether core assumptions are self-consistent
+- Identify logical contradictions
+
+### Phase 2: System Audit
+- Check whether interactions between components are correct
+- Verify data flow and control flow
+- Identify design flaws
+
+### Phase 3: Via Negativa
+- Systematically eliminate "impossible" options
+- Approach truth by eliminating wrong paths
+- Verify satisfaction of necessary conditions
+
+## Output Format
+
+### Audit Report
+
+**Audit Objective**: [Clear audit target]
+
+**Axiom Verification**:
+- [Axiom 1]: [Verification result]
+- [Axiom 2]: [Verification result]
+
+**System Audit**:
+- [Component A]: [Issues found]
+- [Component B]: [Issues found]
+
+**Via Negativa**:
+- Excluded hypothesis 1: [Why impossible]
+- Excluded hypothesis 2: [Why impossible]
+
+**Audit Conclusion**: [Comprehensive judgment]
+
+**Risk Rating**: Low|Medium|High
+
+## Notes
+
+- Each audit point should have a clear basis for judgment
+- Do not skip reasoning steps, proceed gradually
+- If information is insufficient, clearly state what is needed
+- Conclusions should be verifiable, not intuitive judgments
+
+---
+
+Please follow this framework to conduct the audit and output a structured verification report.

package/plugin/templates/langs/en/skills/pd-cli-operator/SKILL.md

@@ -0,0 +1,67 @@
+---
+name: pd-cli-operator
+description: Use this skill when operating the Principles Disciple `pd` CLI: probing runtime health, manually recording pain, inspecting tasks/runs/candidates/artifacts, intaking candidates, viewing flow policy, or cleaning legacy state. It routes CLI commands safely and forbids direct `.state/.pain_flag` writes.
+disable-model-invocation: false
+---
+
+# PD CLI Operator
+
+Use the `pd` CLI as the supported operator surface. Do not modify PD state files by hand unless a command explicitly says to do so.
+
+## Runtime V2 Pain Diagnosis
+
+Manual pain diagnosis:
+```bash
+pd pain record --reason "<reason>" --score <0-100> --workspace "<workspace>" --json
+```
+
+Success requires:
+- `status` is `succeeded`
+- `candidateIds` is non-empty
+- `ledgerEntryIds` is non-empty
+
+Forbidden:
+- Do not write `.state/.pain_flag`.
+- Do not use `write_file`, shell redirection, `Set-Content`, `Out-File`, or `node -e` to create pain flags.
+
+## Health And Policy
+
+Runtime probe:
+```bash
+pd runtime probe --runtime pi-ai --workspace "<workspace>" --json
+```
+
+Flow policy:
+```bash
+pd runtime flow show --workspace "<workspace>" --json
+```
+
+## Inspect Runtime V2 Objects
+
+Tasks and runs:
+```bash
+pd task show --task-id "<taskId>" --json
+pd run show --run-id "<runId>" --json
+```
+
+Candidates and artifacts:
+```bash
+pd candidate list --workspace "<workspace>" --json
+pd candidate show --candidate-id "<candidateId>" --workspace "<workspace>" --json
+pd artifact show --artifact-id "<artifactId>" --workspace "<workspace>" --json
+```
+
+Manual intake:
+```bash
+pd candidate intake --candidate-id "<candidateId>" --workspace "<workspace>" --json
+```
+
+## Legacy Admin
+
+Only when explicitly cleaning old state:
+```bash
+pd legacy cleanup --workspace "<workspace>" --dry-run
+pd legacy cleanup --workspace "<workspace>" --apply
+```
+
+Never use legacy cleanup as a diagnosis trigger.