AgentSkillsCN

analyze

深入分析与调查。通过传递--codex参数，将任务委托给Codex CLI。

SKILL.md
--- frontmatter
name: analyze
description: "Deep analysis and investigation. Pass --codex to delegate to Codex CLI."
argument-hint: "[--codex] [investigation target or question]"

Deep Analysis & Investigation

Execute a thorough analysis using Claude's native analysis capabilities, or delegate to Codex CLI.

Argument Routing

ArgumentMode
<prompt>Claude-native (default)
--codexCodex delegation (context from conversation)
--codex <prompt>Codex delegation

Strip the --codex flag before passing the prompt to the execution path.

Claude-native Execution (default)

  1. •Load protocol: Read agents/analyst.md to load the full analyst protocol
  2. •Apply Investigation_Protocol: Follow the numbered steps in the protocol's <Investigation_Protocol> section
  3. •Use Claude-native tools: Read, Grep, Glob, LSP tools to trace code paths and gather evidence
  4. •Present findings: Use the protocol's <Output_Format> to present findings by severity
  5. •Prioritize: Critical gaps first, nice-to-haves last

Codex Delegation

  1. •Load protocol: Read agents/codex-proxy.md for the prompt template and system instructions. Use the analyst role's prompt template (### Role: analyst section). You call Codex directly - do NOT spawn a codex-proxy agent.
  2. •Gather context: From the conversation, collect:
    • •Investigation target and specific question
    • •File paths, error messages, stack traces
    • •What has been tried or ruled out
  3. •Call Codex: Use codex({ op: "exec", ... }) directly, following the protocol's prompt template. Pass working_directory and reasoning_effort: "xhigh".
  4. •Post-process the raw Codex response:
    • •Verify references: For CRITICAL/HIGH findings, Read the cited file:line to confirm accuracy. Drop findings with incorrect references.
    • •Filter: Remove findings unrelated to the user's question
    • •Restructure: Present findings ordered by severity with verified references
    • •Synthesize: Add a 2-3 sentence summary connecting findings to the original question

Context Enhancement

From the current conversation, identify and include in your analysis:

  • •Error messages, stack traces, or symptoms being investigated
  • •Files and code paths relevant to the issue
  • •What has already been tried or ruled out

Error Policy

If agents/analyst.md (Claude-native) or agents/codex-proxy.md (Codex delegation) cannot be read, report the error to the user. Do not fall back to inline analysis - the agent protocol is a required dependency.