菜单
Assemble Cognitive Context Bundles (CCBs) in parallel for all files in a planned edit set. Call before starting a multi-file editing session to pre-load cognitive load scores and edit precautions.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Run the 5-step Cognitive Analysis Loop to classify files as Pattern A/B/C and emit a per-package cognitive load heatmap. Pass a package path or file glob as the argument.
Converts a feature discussion into a single GitHub Issue for the L0 agent queue using a 5-phase iterative pipeline: (1) generate proposal draft, (2) add implementation details, (3) iteratively review proposal until APPROVED (max 3 attempts), (4) iteratively review goal-phase alignment until APPROVED (max 3 attempts), (5) create GitHub Issue without execution labels for human confirmation. No working-tree switching or remote-push operations.
Discover repository-specific project semantics and write them as `ProjectSemanticsInput` into `.archguard/project-semantics.json` or `archguard.config.json.projectSemantics`. Use when the task is to identify test discovery conventions, assertion wrapper patterns, Mermaid package-layer grouping, or other evidence-backed semantics that ArchGuard should consume.
Execute the full feature development lifecycle for a new feature or substantial change: generate a proposal from discussion or requirements, review and refine the proposal, generate and review an implementation plan, then implement the plan with TDD and validate the result. Use when given a feature discussion, requirements, or existing proposal/plan documents that should be turned into reviewed design docs and working code.
RLM (Refactoring Lifecycle Management): Systematic methodology for managing the complete lifecycle of software refactoring initiatives through 6 phases (PROPOSAL → PLANNING → EXECUTION → VALIDATION → INTEGRATION → MONITORING) with bi-directional traceability, quality gates, and metrics-driven measurement. Use when managing multiple refactoring iterations, improving plan-execution alignment, establishing refactoring dashboard and traceability matrix, or coordinating team refactoring efforts.
| name | cognitive-prep |
| description | Assemble Cognitive Context Bundles (CCBs) in parallel for all files in a planned edit set. Call before starting a multi-file editing session to pre-load cognitive load scores and edit precautions. |
| argument-hint | <space-separated file paths> |
| allowed-tools | ["mcp__archguard__archguard_get_ccb","Read"] |
Before editing multiple files, assemble their CCBs in parallel so the agent enters each edit with full cognitive context pre-loaded.
List target files: parse the argument as space-separated file paths (or use the current task's planned edit files if no argument given).
Assemble CCBs in parallel: for each file, call archguard_get_ccb with { filePath: "<path>" }. Run all calls concurrently.
Display summary table:
| File | Pattern | Cognitive Load | Top Edit Precaution |
|---|---|---|---|
| ... | A/B/C | 0.0–1.0 | ... |
Highlight high-risk files: any file with cognitiveLoad >= 0.7 or Pattern B — flag with WARNING and display all editPrecautions.
Proceed: the agent now has full CCB context for each planned edit and can begin work with minimal redundant exploration.
