Coverage for mcp_bridge/tools/templates.py: 0%

1""" 

2Templates for stravinsky repository initialization. 

3""" 

4 

5CLAUDE_MD_TEMPLATE = """## stravinsky MCP (Multi-Model Orchestration) 

6 

7Stravinsky provides multi-model AI orchestration with parallel agent execution. 

8 

9### Architecture 

10- **Native Subagent**: Stravinsky orchestrator (.claude/agents/stravinsky.md) auto-delegates complex tasks 

11- **MCP Tools**: agent_spawn, invoke_gemini, invoke_openai, LSP tools, code search 

12- **Specialist Agents**: explore, dewey, frontend, delphi, multimodal, document_writer 

13 

14### Agent Tools (via MCP) 

15- `agent_spawn(prompt, agent_type, description)` - Spawn background agent with full tool access 

16- `agent_output(task_id, block)` - Get results (block=True to wait) 

17- `agent_progress(task_id)` - Check real-time progress 

18- `agent_list()` - Overview of all running agents 

19- `agent_cancel(task_id)` - Stop a running agent 

20 

21### Agent Types 

22- `explore` - Codebase search, structural analysis (Gemini 3 Flash) 

23- `dewey` - Documentation research, web search (Gemini 3 Flash + Web) 

24- `frontend` - UI/UX implementation (Gemini 3 Pro High) 

25- `delphi` - Strategic advice, architecture review (GPT-5.2 Medium) 

26- `multimodal` - Visual analysis, screenshots (Gemini 3 Flash Vision) 

27- `document_writer` - Technical documentation (Gemini 3 Flash) 

28 

29### Parallel Execution (MANDATORY) 

30For ANY task with 2+ independent steps: 

311. **Immediately use agent_spawn** for each independent component 

322. Fire all agents simultaneously in ONE response, don't wait 

333. Monitor with agent_progress, collect with agent_output 

34 

35### Trigger Commands 

36- **IRONSTAR** / **IRS**: Maximum parallel execution - spawn agents aggressively for every subtask 

37- **ULTRATHINK**: Engage exhaustive deep reasoning, multi-dimensional analysis 

38- **SEARCH**: Maximize search effort across codebase and external resources 

39- **ANALYZE**: Deep analysis mode with delphi consultation for complex issues 

40 

41### Native Subagent Benefits 

42- ✅ Auto-delegation (no manual /stravinsky invocation) 

43- ✅ Context isolation (orchestrator runs as subagent) 

44- ✅ Full MCP tool access (agent_spawn, invoke_gemini/openai, LSP, etc.) 

45- ✅ Multi-model routing (Gemini for UI/research, GPT for strategy) 

46""" 

47 

48COMMAND_STRAVINSKY = """--- 

49description: stravinsky Orchestrator - Parallel agent execution for complex workflows. 

50--- 

51 

52## CRITICAL: USE STRAVINSKY MCP TOOLS 

53 

54You MUST use the Stravinsky MCP server tools for ALL file reading, searching, and parallel work. 

55 

56### MANDATORY TOOL USAGE: 

57 

58**For ANY file reading or searching:** 

59``` 

60stravinsky:agent_spawn( 

61 prompt="Read and analyze [file path]. Return: [what you need]", 

62 agent_type="explore", 

63 description="Read [file]" 

64) 

65``` 

66 

67**For documentation/library research:** 

68``` 

69stravinsky:agent_spawn( 

70 prompt="Find documentation for [topic]", 

71 agent_type="dewey", 

72 description="Research [topic]" 

73) 

74``` 

75 

76**For 2+ independent tasks (ALWAYS parallel):** 

77``` 

78// Fire ALL at once in ONE response - NEVER sequential 

79stravinsky:agent_spawn(prompt="Task 1...", agent_type="explore", description="Task 1") 

80stravinsky:agent_spawn(prompt="Task 2...", agent_type="explore", description="Task 2") 

81stravinsky:agent_spawn(prompt="Task 3...", agent_type="dewey", description="Task 3") 

82// Then immediately continue - don't wait 

83``` 

84 

85**To get results:** 

86``` 

87stravinsky:agent_output(task_id="[id]", block=true) 

88``` 

89 

90### Recommended Tool Usage: 

91- For file operations within agents: Use standard Read/Edit tools 

92- For parallel agent spawning: Use stravinsky:agent_spawn (supports nesting, unlike native Task tool) 

93- For collecting results: Use stravinsky:agent_output 

94- For monitoring agents: Use stravinsky:agent_list 

95 

96### Native Subagent Integration: 

97- Stravinsky orchestrator configured as native Claude Code subagent (.claude/agents/stravinsky.md) 

98- Native subagents CAN call Stravinsky MCP tools (agent_spawn, invoke_gemini, etc.) 

99- This enables auto-delegation without manual /stravinsky invocation 

100 

101### Execution Modes: 

102- `ironstar` / `irs` - Maximum parallel execution (10+ agents) 

103- `ultrathink` - Deep reasoning with delphi consultation 

104- `search` - Exhaustive multi-agent search 

105 

106**Your FIRST action must be spawning agents, not using Read/Search tools.** 

107""" 

108 

109COMMAND_PARALLEL = """--- 

110description: Execute a task with multiple parallel agents for speed. 

111--- 

112 

113Use the stravinsky MCP tools to execute this task with PARALLEL AGENTS. 

114 

115**MANDATORY:** For the following task items, spawn a SEPARATE `agent_spawn` call for EACH independent item. Do not work on them sequentially - fire all agents simultaneously: 

116 

117$ARGUMENTS 

118 

119After spawning all agents: 

1201. Use `agent_list` to show running agents 

1212. Use `agent_progress(task_id)` to monitor each 

1223. Collect results with `agent_output(task_id, block=True)` when ready 

123""" 

124 

125COMMAND_CONTEXT = """--- 

126description: Refresh project situational awareness (Git, Rules, Top Todos). 

127--- 

128 

129Call the `get_project_context` tool to retrieve the current Git branch, modified files, local project rules from `.claude/rules/`, and any pending `[ ]` todos in the current scope. 

130""" 

131 

132COMMAND_HEALTH = """--- 

133description: Perform a comprehensive system health and dependency check. 

134--- 

135 

136Call the `get_system_health` tool to verify that all CLI dependencies (rg, fd, sg, tsc, etc.) are installed and that authentication for Gemini and OpenAI is active. 

137""" 

138 

139COMMAND_DELPHI = """--- 

140description: Consult the delphi Strategic Advisor for architecture and hard debugging. 

141--- 

142 

143Use the `delphi` prompt to analyze the current problem. This triggers a GPT-based consulting phase focused on strategic reasoning, architectural trade-offs, and root-cause analysis for difficult bugs. 

144 

145**When to use delphi:** 

146- Complex architecture design decisions 

147- After 2+ failed fix attempts 

148- Multi-system tradeoffs 

149- Security/performance concerns 

150- Unfamiliar code patterns 

151""" 

152 

153COMMAND_LIST = """--- 

154description: List all active and recent background agent tasks. 

155--- 

156 

157Call the `agent_list` tool to see an overview of all currently running and completed background agents, including their Task IDs and statuses. 

158""" 

159 

160COMMAND_DEWEY = """--- 

161description: Trigger dewey for documentation research and implementation examples. 

162--- 

163 

164Use the `dewey` prompt to find evidence and documentation for the topic at hand. dewey specializes in multi-repository search and official documentation retrieval. 

165 

166**When to use dewey:** 

167- Unfamiliar npm/pip/cargo packages 

168- "How do I use [library]?" 

169- "What's the best practice for [framework feature]?" 

170- Finding OSS implementation examples 

171""" 

172 

173COMMAND_VERSION = """--- 

174description: Returns the current version and diagnostic info for stravinsky. 

175--- 

176 

177Display the stravinsky MCP version, registered hooks, available agents, and system health status. 

178""" 

179 

180SLASH_COMMANDS = { 

181 "stravinsky.md": COMMAND_STRAVINSKY, 

182 "parallel.md": COMMAND_PARALLEL, 

183 "list.md": COMMAND_LIST, 

184 "context.md": COMMAND_CONTEXT, 

185 "health.md": COMMAND_HEALTH, 

186 "delphi.md": COMMAND_DELPHI, 

187 "dewey.md": COMMAND_DEWEY, 

188 "version.md": COMMAND_VERSION, 

189}