Popular repositories Loading

1. harness-loop harness-loop Public

   Minimal self-improving agent harness: run → mine failure traces → regression-gate → inject rules. Companion repo to a Cloud Codex blog post.

   Python 1

2. agent-loop agent-loop Public

   A bounded agent loop with three hard exits — iteration cap, cost ceiling, and stall detection.

   TypeScript

3. inference-router inference-router Public

   Provider-agnostic LLM inference client — swap backends with an env var, with a built-in smoke test.

   TypeScript

4. mcp-token-audit mcp-token-audit Public

   Measure the token cost of MCP server tool schemas and generate a defer_loading manifest.

   Python

5. token-baseline token-baseline Public

   Measure a prompt/skill/command corpus's token footprint, track reductions across snapshots, find duplicate files.

   Python

6. ballast ballast Public

   An agent reliability layer on OpenTelemetry — prompt-contracts, guardrails, bounded loops, and eval as ballast.* extensions on gen_ai.* spans. OTel tells you what happened; ballast tells you whethe…

   TypeScript