SkillHarvest – Turn agent corrections into reusable skills

Reads local coding-agent sessions and converts user corrections into reusable skills so agents stop repeating mistakes.

Problem

Developers use multiple coding agents like Claude Code, Codex, and Cursor, but corrections to AI mistakes stay in the current session and agents repeat the same errors next time. Existing tools either don't read local sessions or require manual tidying, lacking automation. Users also demand strong privacy and consent controls.

Solution

SkillHarvest reads existing local coding-agent session logs, automatically identifies where users corrected the agent, and extracts structured skills (e.g., 'this project doesn't use any type'), which agents load next time via a skill mechanism. Every capture source is off by default with its own consent setting.

Users

Professional developers using multiple coding agents and heavy AI-assisted programming users.

MVP

Scan local session files for Claude Code, Codex, Cursor; auto-detect correction patterns and generate skill drafts; write confirmed skills into agent-loadable skill directories; independent toggles and consent per capture source; export to universal skill format.

Revenue

Open-source core + team subscription ($8/user/mo) for shared skill libraries and cross-member sync; personal one-time purchase at $25.

Why now

Coding agent usage is exploding, but each agent's memory is isolated, making repeated corrections costly. Local session data already exists and can be mined with a lightweight tool, while privacy compliance makes local-first more appealing.