Define requirements for an OpenClaw integration
Document a proposed integration's data needs, actions, credentials, and review boundaries.
Deep analysis mode. Gather comprehensive context before diving deep. Use for investigation, debugging, and understanding complex systems.
--- name: analyze-mode-b7c2d15c9fb8 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: ``` // 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: | Tool | Purpose | |------|---------| | `search_semantic` | Natural language code search | | `find_similar` | Find similar code patterns | | `find_dependencies` | What depends on X? | | `call_graph` | Function caller/callee relationships | | `impact_analysis` | Change risk assessment | | `grep` | Find specific patterns | | `ast_grep_search` | Structural analysis | | `lsp_goto_definition` | Jump to symbol definition | | `lsp_find_references` | Find all usages | | `lsp_symbols` | Document/workspace symbol search | | `lsp_diagnostics` | Type errors before build | | `git log -S` | History evolution | | `git blame` | Change 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 ``` 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] ```--- name: analyze-mode-b7c2d15c9fb8 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: ``` // 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: | Tool | Purpose | |------|---------| | `search_semantic` | Natural language code search | | `find_similar` | Find similar code patterns | | `find_dependencies` | What depends on X? | | `call_graph` | Function caller/callee relationships | | `impact_analysis` | Change risk assessment | | `grep` | Find specific patterns | | `ast_grep_search` | Structural analysis | | `lsp_goto_definition` | Jump to symbol definition | | `lsp_find_references` | Find all usages | | `lsp_symbols` | Document/workspace symbol search | | `lsp_diagnostics` | Type errors before build | | `git log -S` | History evolution | | `git blame` | Change 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 ``` 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] ```
Imported from the listed AgentSkill.sh category. Complete supplied package files and notices are retained. Required external tools and credentials are not installed. No bundled code was executed during import. Individual agent task behavior has not been tested.