4lbert Logo~/4lbert
HomeProfileSkillsPortfolioBlogGalleryContact
$ echo "© 2026 Torsten Albert · 4lbert.com"
ImpressumDatenschutz
Back to blog

Complexity of Composition Problem in AI Coding Agents

TorstenMarch 26, 20262 min read

Secret-Service is a Python MCP server that routes problems through a pipeline of seven specialised LLM-backed agents using a blackboard architecture stored entirely in SQLite. Instead of committing to one solution, it fans out into parallel strategy branches, gates each plan through a Judge before execution, scores the results, and returns the winner. Over time, the system learns: successful strategies are embedded and recalled for future problems; failures become anti-patterns.

On this page
  • Problem
  • What It Is
  • Architecture
  • The 7 Agents
  • Memory System
  • Observability
  • Scope Limits
  • Quick Start
  • Stats

Secret-Service: Multi-Agent MCP Server for Structured Problem-Solving

Problem

Debugging across composed LLM tool layers (skills, hooks, subagents, MCP servers) lacks observability and traceability. Individual pieces work; their interactions don't.

What It Is

A Python MCP server that routes problems through 7 specialised LLM agents in a structured pipeline. Multiple strategies compete in parallel; the best wins. Every step is a structured, queryable event. Runs fully local — SQLite + sqlite-vss for storage/vector search, sentence-transformers for embeddings, MCP sampling for LLM inference. Zero external dependencies, zero API keys.

Architecture

Blackboard pattern: All agents read/write to a shared SQLite DB — no hidden state. Parallel fan-out: The Strategist generates N strategies (default 3), each executed concurrently in its own branch.

Pipeline (14 LLM calls for 3 strategies):

  1. Intake (sequential): Reception → Master → Strategist
  2. Execution (parallel per strategy): Taktik Planner → Judge → Mission (Judge rejects? Retry up to 3×)
  3. Evaluation (sequential): Jury scores all Missions → Master synthesises final answer

The 7 Agents

  • Reception (temp 0.1) — precise problem intake
  • Master (0.3) — orchestration and synthesis
  • Strategist (0.9) — divergent strategy generation
  • Taktik Planner (0.8) — creative step planning
  • Judge (0.1) — rigorous plan verification (gatekeeper before execution)
  • Mission (0.2) — faithful plan execution
  • Jury (0.2) — consistent multi-dimensional scoring

    • High temp = creativity. Low temp = quality control.

    Memory System

    Learns from every session:

    • High-scoring successes → good_practice + pattern memories (permanent, reusable)
    • Failures → bad_practice / anti_pattern memories
    • Memories are embedded and recalled via similarity search for future strategy generation
    • Integrity: confidence decay on contradictions, near-duplicate supersession, relevance tracking

    Observability

    Append-only structured event stream (16 event types) with typed payloads. inspect_session() returns the full causal chain from problem to solution. No log grepping.

    Scope Limits

    • Does not write code, run tests, or modify files — returns structured recommendations
    • Does not call LLM providers directly — requires MCP sampling support from the client
    • Does not replace human judgment

    Quick Start

    git clone <repo-url> && cd Secret-Service
pip install -e ".[dev]"

    Workflow: solve() → poll get_events() → get_result() → optionally inspect_session().

    Stats

    7 agents · 14 LLM calls/session · 13 DB tables · 8 MCP tools · 239 tests · 0 external dependencies · ~2,500 lines of Python.

    https://github.com/TorstenAlbert/secret-service

    PreviousDevAgentOps: Workflow for LLM Agents in CI/CDNextSpin the wheel

    On this page

    • Problem
    • What It Is
    • Architecture
    • The 7 Agents
    • Memory System
    • Observability
    • Scope Limits
    • Quick Start
    • Stats