Amadeus Skill

• The description states that Amadeus is a local-first control plane for procedural memory.
• It captures opt-in, session-scoped Codex runs, distills verified work into Agent Skills or deterministic scripts, and governs their lifecycle through review and promotion gates.
• The system uses:
  • A Codex CLI (no API key needed).
  • SQLite + FTS5 for local state and retrieval.
  • A Fastify dashboard server with React frontend.
  • An MCP stdio server connecting Codex sessions to Amadeus evidence.
  • Agent Skills packages with SKILL.md and agents/openai.yaml.
• It is described as a single npm-distributable TypeScript product.
• The system does not activate skills automatically; all candidate packages must be reviewed and approved by humans before promotion.

Note: This is self-reported. No evidence of actual customers, revenue or usage beyond internal testing.

• The description states that Amadeus is not another prompt library nor long-term chat memory.
• It positions itself as:
  • Outcome-first (grades whether the task was verified).
  • Reliability-first (must pass capability, safety, and economic gates).
  • Value-aware (procedures must improve or retain success with material efficiency).
  • Human-governed (candidate creation, testing, and promotion are explicit human decisions).
  • Local-first (data stays local).
• The authors claim that Amadeus is not a tool for saving every successful answer, but instead focuses on reusable knowledge that has been validated.
• It evolved from a demo to a system that tests real-world code migration and race condition fixes, demonstrating a shift from "what works" to "what should be reused".

Inference: The positioning reflects an attempt to differentiate from generic agent libraries by emphasizing governance, safety, and value-awareness.

• Not evidenced.
• The description does not name specific customer types or personas.
• It implies use by developers working with coding agents (e.g., Codex), but no explicit target segment is defined.

Absence of evidence: No clear indication of who the intended users are beyond developers using Codex.

• Not evidenced.
• The description does not mention any pricing, monetization strategy, or business model.
• It describes a local-first tool with no API key requirement and an npm package distribution mechanism.

Absence of evidence: No indication of how the product would be sold or whether it’s intended for commercial use.

• Built using:
  • TypeScript
  • Node.js
  • Fastify (local API server)
  • React (dashboard UI)
  • SQLite + FTS5 (local storage and retrieval)
  • Codex CLI
  • MCP stdio server
  • npm package workflow for distribution
• The system is described as:
  • A single npm-distributable TypeScript product
  • Uses native Codex CLI with existing ChatGPT login
  • Does not require OpenAI API keys
  • Has a local-first architecture
  • Supports deterministic similarity, safe regex filters, and abstention logic
• The authors mention:
  • A machine-verified v1.1 implementation
  • Manual exercise from installation through controlled evaluation
  • Use of MCP receipts to verify candidate ID, content hash, and prepared path
  • Bounded Codex process-group timeout for safety

Inference: The technical stack suggests a developer-focused tool built for local use with strong emphasis on control and auditability.

• Not evidenced.
• The description states:
  • No current candidate has passed every promotion gate.
  • Active skill count remains zero.
  • The npm tarball works for private distribution; registry publication is pending.
  • All three governed candidates were blocked due to economic regressions or value issues.
• There is no mention of:
  • Customers
  • Revenue
  • Usage metrics
  • Product adoption beyond internal testing

Absence of evidence: No signs of traction, adoption, or commercial use.

• Not evidenced.
• The description does not name competitors or reference existing tools in the space.
• It implies Amadeus is distinct from prompt libraries and long-term memory systems.
• It references Codex as the first harness adapter but does not discuss other agent platforms or tools.

Absence of evidence: No competitive landscape or positioning relative to similar tools.

• No commercial traction: The system has not been promoted beyond internal testing and no skills are active.
• Limited scope: Only Codex is supported as a harness; no other agent platforms are mentioned.
• Self-reported maturity: The project is described as a hackathon submission with no external validation or user feedback.
• No monetization strategy: No indication of how the product would be sold or scaled.
• High human governance overhead: Every skill must go through multiple gates and human approvals, which may limit scalability.
• Local-first design may limit adoption: A local-only system might not appeal to teams seeking centralized control or collaboration features.

Inference: The project is in early development with no evidence of real-world usage or commercial viability.

1. What specific use cases have you identified for Amadeus beyond the internal testing?
2. How do you plan to scale governance without human bottlenecks?
3. Are there any plans to support other agent platforms beyond Codex?
4. What is your roadmap for monetization or commercial deployment?
5. Have you tested Amadeus in production environments with real teams?
6. How do you handle edge cases where a skill fails after promotion?
7. What are the key metrics you track to evaluate whether a skill should be promoted?