harvest-loop - Harvest field sessions into FH evolution
Runs an eight-step pipeline that extracts session learnings, checks skill conflicts, evaluates candidates, and records or proposes FH ecosystem changes.
Tags
Updated: 2026-10-03Capabilities
Typical Inputs
Typical Outputs
What this skill does
- Detect session regressions
- Extract field patterns
- Compare patterns with skills
- Challenge existing skills
- Propose skill candidates
- Synthesize and grade proposals
- Critique proposals independently
- Check harness health
- Verify bidirectional consistency
- Curate lifecycle changes
Inputs
- Field-session git diff
- Session outputs
- Existing FH skills
- Git history
- Reference session cards
- Memory entries
- User approval decision
- Adaptation profile
- Local operator configuration
Outputs
- Harvested pattern summary
- Final proposal list
- fh_signal files
- Verdict line
- Pull request
- Skill lifecycle updates
- Execution confirmation prompt
- No-targets message
Requirements
- Git-based FH workspace
- Read and write access
- Shell execution access
- Read, Write, Bash, Grep, Glob, and Agent tools
- Agent-capable model