ingot

Overview

Ingot acts as an orchestration and version-control layer for AI agent skills, integrating with optimizer tools like SkillOpt to train, test, and safely promote instruction edits. It quarantines optimizer-generated challengers, requires evidence-based verification against held-out tasks, and provides a local UI and audit trail for human-in-the-loop promotion, rollback, and management.

Capabilities

• ▸Content-addressed skill revisions with cryptographic hashing
• ▸Quarantined execution and storage for optimizer challengers
• ▸Evidence-gated promotion workflows comparing champion vs challenger performance
• ▸Staged and recoverable skill installation with automatic rollback
• ▸Model Context Protocol (MCP) server for serving skill content
• ▸Integrated Langfuse telemetry and trace mining
• ▸Web-based dashboard for human review, approval, and diff inspection

Best for

Managing and versioning AI agent instruction files (skills), Evaluating and promoting optimizer-generated skill improvements with human oversight, Tracing agent execution and routing behavior via integrated Langfuse observability, Connecting external coding agents like Claude Code or Codex to a controlled skill registry via MCP

Works with

Security considerations

The MCP endpoint does not feature built-in authentication, and Ingot is not designed as a hardened multi-tenant service. Exposing it beyond a trusted machine requires following production security setup guides.

Categories