learning-capture

name	learning-capture
description	Conditionally captures verified reusable ADLC learnings into docs/solutions after successful closeout.
contract_version	1.0.0
side_effect_profile	mutating
activation	{"mode":"conditional_closeout","consumes_manifest":true,"trigger_nodes":["pr_prep","learning_capture"],"produces":["learning_candidates","learning_entry"]}

Learning Capture

Purpose

Capture only reusable, verified learnings that make future ADLC runs cheaper or safer. This is a closeout path, not a planning replacement and not a default documentation chore.

Trigger

Run after pr_prep when the PR package includes learning_candidates.

Skip when:

• the work was mechanical, lint-only, or one-off
• tests did not pass
• the candidate lacks source evidence or verifier evidence
• the content may include secrets, private tokens, credentials, or unsupported environment claims
• the candidate duplicates an existing docs/solutions entry without adding new evidence

Process

1. Read only the candidate, the cited source evidence, and the nearest relevant existing docs/solutions entries.
2. Choose create, update, or skip.
3. Write one compact markdown entry using docs/solutions/_template.md.
4. Include source_evidence, verifier, redaction_review, and stale_conditions.
5. If the candidate came from a deviation log, include whether it was a brief-generator defect, which epistemic ledger entry was absent or stale, and how future briefs should avoid the ledger-absent decision.
6. If the candidate came from a completion audit or Minimalism Auditor finding, include the predicate-library rule ID, the violated detection predicate, and whether the finding was thinness or an honesty-contract violation.
7. Run python3 scripts/validate_learning_entry.py <entry>.
8. Write the closeout decision JSON to ADLC process artifact storage with bin/adlc process-artifact-path --target-repo <target-repo> --task <task-id> --artifact-type closeout --filename learning-capture.json --json.
9. Emit pass only when validation passes; emit skipped when there is no reusable verified learning.

Output

{
  "label": "pass | skipped | fail",
  "learning_capture": {
    "action": "create | update | skip",
    "path": "docs/solutions/slug.md | null",
    "reason": "string",
    "verifier": "python3 scripts/validate_learning_entry.py docs/solutions/slug.md",
    "closeout_artifact_path": "string",
    "redaction_status": "passed | needs_review"
  }
}

Rules

• Do not capture secrets or raw local credentials.
• Do not capture unsupported architecture claims.
• Do not capture a ledger-absent deviation as a verified learning unless the missing epistemic ledger evidence and correction are explicit.
• Do not paste entire session transcripts.
• Do not create broad refresh work; emit stale conditions for learning-refresh.
• A learning entry is prior art, not proof. Future agents must still verify current behavior.