Record agent tool-call trajectories to SQLite, replay against a different model/prompt for cheap A/B testing. Measure planning quality without burning fresh API calls.
npx add-skill fr3kchy/trajectory-replay-demoRecord agent tool-call trajectories to SQLite, replay against a different model/prompt for cheap A/B testing. Measure planning quality without burning fresh API calls.
Source: https://github.com/fr3kchy/trajectory-replay-demo
Discovered from GitHub repositories pushed in the last 24 hours for agent skills, Claude/Codex/Gemini workflows, MCP tooling, and adjacent AI-agent automation.
A skill that installs Ralph Wiggum — an autonomous AI agent loop that works through your GitHub Issues unattended. Heavily inspired by the work of Matt Pocock.
Claude Code skill for analysing the trajectory and reporting over usefullness of tools
Agent skills for fallow, codebase intelligence for JavaScript and TypeScript. Teaches AI agents how to find unused code, duplication, circular deps, complexity hotspots, architecture drift, and (with Fallow Runtime) hot-path and cold-path evidence. Works with Claude Code, Cursor, Codex, Gemini CLI, and 30+ agents.
Big Red Button — the universal autonomous agent. Cowork × Manus × Perplexity × Lovable. Submitted to Runway 2026 API Hackathon.
Run local code as if it were a pod in a remote Kubernetes cluster: real env vars, DNS, network, traffic. Used by AI coding agents and developers.
A predictable hub for local and subscribed AI agent skills