By name: h2-harvest description: Curate staged Markdown or text notes from docs/_harvest-inbox into canonical docs using the harness h2-harvest CLI. Use when the user asks to harvest inbox notes, promote/reject a staged knowledge file, or inspect h2-harvest behavior.
h2-harvest
Use h2-harvest for inbox-based knowledge curation outside the feature lifecycle.
Runtime command:
.harness-helm/scripts/harness h2-harvest [--promote <path> | --reject <path>] [--dry-run] [--force] [--skip-raw]
Inputs must live under docs/_harvest-inbox/{solution,convention,domain,spec,decision,ops}/ and must be .md or .txt. .txt inputs are written to canonical destinations as .md.
If docs/_harvest-inbox/raw/ has prefix-free files, h2-harvest prints a warning and suggests running h2-harvest-tag first. Use --skip-raw only when intentionally ignoring raw inbox files.
Phase 1 does not use LLM or embedding classification. It applies deterministic policy from .harness-helm/h2-harvest.yml, validates frontmatter, warns on possible semantic type mismatches, and keeps governed destinations pending unless explicitly approved by the workflow.
Low-risk solution and convention entries can auto-write only when confidence: high and evidence metadata are present. Governed entries (domain, spec, decision, ops) remain pending. --reject or status: rejected hard-deletes the inbox source after writing the run report; use --dry-run to preview.
Reports are written to .harness-helm/runs/_unscoped/{run-id}/harvest-report.md.