AgentSkillsCN

analyze-mode

深度分析模式。在深入探究之前,先全面收集背景信息。适用于调查、调试,以及理解复杂系统。

SKILL.md
--- frontmatter
name: analyze-mode
description: Deep analysis mode. Gather comprehensive context before diving deep. Use for investigation, debugging, and understanding complex systems.

Analyze Mode

ACTIVATION: When loaded, gather comprehensive context before any deep analysis.

Protocol

ANALYSIS MODE - Context First, Conclusions Second

Phase 1: Context Gathering (Parallel)

Deploy 2-4 agents for comprehensive context:

code
// Internal context (1-2 agents)
task(agent="explore", prompt="Find all implementations of [topic]", background=true)
task(agent="explore", prompt="Find tests and usage patterns for [topic]", background=true)

// External context (if libraries involved)
task(agent="researcher", prompt="Find official docs for [library]", background=true)
task(agent="researcher", prompt="Find known issues and solutions", background=true)

Phase 2: Direct Analysis

While agents run, use direct tools:

ToolPurpose
search_semanticNatural language code search
find_similarFind similar code patterns
find_dependenciesWhat depends on X?
call_graphFunction caller/callee relationships
impact_analysisChange risk assessment
grepFind specific patterns
ast_grep_searchStructural analysis
lsp_goto_definitionJump to symbol definition
lsp_find_referencesFind all usages
lsp_symbolsDocument/workspace symbol search
lsp_diagnosticsType errors before build
git log -SHistory evolution
git blameChange attribution

Phase 3: Oracle Consultation (If Complex)

Escalate to Oracle when:

  • Architecture spans multiple systems
  • Debugging has failed 2+ times
  • Trade-off analysis needed
  • Root cause remains unclear
code
task(agent="oracle", prompt="Analyze [problem] given context...")

Analysis Framework

For Debugging

  1. Reproduce - Can you trigger the issue?
  2. Isolate - Where exactly does it fail?
  3. Hypothesize - What could cause this?
  4. Test - Verify each hypothesis
  5. Fix - Address root cause, not symptom

For Investigation

  1. Scope - What are the boundaries?
  2. Dependencies - What does it depend on?
  3. Dependents - What depends on it?
  4. Patterns - How is it typically used?
  5. Edge Cases - What are the limits?

For Architecture

  1. Current State - What exists today?
  2. Requirements - What must be satisfied?
  3. Options - What approaches are viable?
  4. Trade-offs - What are the costs/benefits?
  5. Recommendation - What should we do?

Output Requirements

Synthesize analysis into:

markdown
## Analysis Summary

**Subject:** [What was analyzed]

**Context Gathered:**
- [Finding 1 from explore/researcher]
- [Finding 2]

**Key Observations:**
- [Observation 1]
- [Observation 2]

**Root Cause / Conclusion:**
[Clear statement]

**Recommended Action:**
[Specific next steps]

**Confidence Level:** [High/Medium/Low] - [Reasoning]