SuperAPI AI Directory

black-tortoise-agent-orchestration

Agents & models

View repository

Updated 10-15 07:54

About this skill

Black-Tortoise Agent Orchestration Skill

When to use

  • Use Black-Tortoise agents to handle multi-step tasks (planning → implementation → verification).
  • Need to coordinate multiple MCPs within a single turn (angular-cli, Software-planning-mcp, sequentialthinking, context7, codacy).

Workflow (minimal)

  1. Locate rules: according to docs/INDEX.md read the corresponding AGENTS.md → README.md。
  2. Break down: call sequentialthinking/* to get steps; if a formal plan is needed, use Software-planning-mcp/* 's start_planning / save_plan。
  3. Verify:
    • API / framework:context7/*(resolve-library-id → get-library-docs)。
    • Angular CLI / schematics:angular-cli/*。
  4. Implementation guardrails:
    • Unidirectional dependency: Presentation → Application → Domain → Infrastructure.
    • Signals-first, prohibit RxJS state management and as any。
    • Maintain deletability, minimal changes.
  5. Quality: when necessary use codacy/* to get analysis, then provide actionable fixes; recommend running pnpm run lint、pnpm run architecture:gate(do not execute automatically).
  6. Sub-agent switching:
    • Planning:agent/runSubagent → Planning mode instructions
    • Debugging:agent/runSubagent → debug
    • Security/quality:agent/runSubagent → SE: Security

MCP anti-loop rules

  • Call the same MCP only once for the same topic; if there is no new information, stop repeating immediately.
  • context7:resolve-library-id followed by get-library-docs complete in one round, avoid re-running.
  • Software-planning-mcp:start_planning / save_plan execute only once when the plan needs updating.
  • sequentialthinking: use for initial breakdown/prioritization, do not re-enter within the same step.
  • Avoid nested or mutually recursive agent/runSubagent ; switch only one sub-agent at a time, and return to the main agent after completion.

Output requirements

  • List assumptions and risks first, then provide concrete steps (including file paths).
  • Point out commands or tests that need verification.
  • Keep it concise, avoid repetition, avoid cross-domain coupling.

Example input

  • "Add list sorting to the tasks capability, give the plan first then implement"
  • "Check the sticky behavior of mat-table in Angular Material 20, provide fix steps"

Example output key points

  • Fragmented steps + path links
  • Specify the AGENTS.md / README.md
  • Remind to run lint / architecture gate / relevant tests