Use compact kind:value labels when recording reviewed task evidence. Labels
should describe what a human or deterministic check verified, not just what
command happened to run.
Recommended labels:
test:<name>: automated test passed, e.g.test:pytestlint:<name>: lint/static check passeddiff:reviewed: human reviewed the diffhealthcheck:<name>: service or runtime health check passedprovider:<request-or-response-id>: external provider result was checkedartifact:<path>: durable artifact was produced and inspectedrollback:<checkpoint-id>: rollback path was verifieddirect:<path-or-id>: direct corpus/file/ledger inspection was performed
Avoid vague labels such as done, ok, or checked because they do not say
what evidence was reviewed.
helm harness record-evidence --completion-evidence and helm task complete --evidence both accept these labels. Recording evidence does not by itself
approve risky follow-up actions.