菜单
Before answering any question or starting any task, identify what files and context are needed for an accurate, complete response. Use when the agent needs to understand what it doesn't know yet — the meta-cognitive 'stop and think' gate.
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
You MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior. Explores user intent, requirements and design before implementation. Produces a validated design doc, then invokes writing-plans.
Generate a comprehensive map of all files relevant to a task before making any changes. Use when about to implement features, fix bugs, or refactor code — maps files to modify, dependencies, tests, reference patterns, and risk assessment.
Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Claude's capabilities with specialized knowledge, workflows, or tool integrations.
TaskLang pseudocode language for AI agent prompting. Use when writing, reviewing, or refactoring agent definitions, skill playbooks, commands, workflow pipelines, or project rules in TaskLang syntax.
Create tools on the fly — Bash scripts, Python scripts, Node.js scripts, or request MCP server/CLI installation — when the agent identifies a capability gap. Use when a task requires a tool that doesn't exist yet, or when an MCP server or CLI tool would solve the problem better than manual steps.
| name | what-context-needed |
| description | Before answering any question or starting any task, identify what files and context are needed for an accurate, complete response. Use when the agent needs to understand what it doesn't know yet — the meta-cognitive 'stop and think' gate. |
TASKLANG TYPE SKILL
IDENTITY "What Context Do I Need?"
Meta-cognitive skill that makes agents self-aware about their knowledge gaps. Before answering any question, the agent identifies what files it needs to see, what it's uncertain about, and what it already knows from the conversation. This is the playbook for someone who asks the right questions before acting.
!!! The #1 failure mode of AI agents is acting without sufficient context. !!! This skill is a GATE — run it BEFORE jumping to answers. !!! Better to ask for 5 files and be right than guess and be wrong.
MAP workflows "workflows/analyze-question.md" -> "Decompose question into sub-questions, identify files needed for each" "workflows/gather-context.md" -> "Systematically read identified files and extract relevant information" "workflows/validate-context.md" -> "Verify gathered context is sufficient before proceeding"
KNOWLEDGE PATTERNS - "Question decomposition — break complex questions into sub-questions, each needing specific files" - "Import tracing — if a file is relevant, its imports and importers may be too" - "Test adjacency — for any source file, check for co-located test files" - "Config awareness — check for .env, config files, package.json that affect behavior" - "Type definition hunting — when code references types, trace to their definition files" - "Already-seen tracking — maintain awareness of what's in the conversation context" - "Glob before Grep — use file patterns to narrow scope before searching content"
CONVENTIONS - "Always categorize: Must See (required) | Should See (helpful) | Already Have | Uncertain" - "Explain WHY each file is needed — not just list paths" - "Check conversation history before requesting files already discussed" - "Prioritize Must See — max 10 files. If more needed, decompose the question" - "When uncertain, say so explicitly with what would resolve it" - "Never answer a question while flagging [GAP] — resolve gaps first"
STRATEGY CAPTURE: patterns of which file types are needed for which question types UPDATE_FREE: references/ UPDATE_APPROVAL: SKILL.md
REQUIRE
