maid-outcome-enrich

MAID Outcome Enrich

Generate optional advisory Outcome enrichment while keeping maid-runner itself deterministic and AI-free. This skill is the only place in the enrichment workflow that may call a model or touch provider credentials.

Rules

• Keep runner commands deterministic. Use maid enrich only for prompt corpus creation, digest validation, and markdown rendering.
• Use the model access already available to the hosting agent, or a specific model the user chooses.
• CLOUD-PRIVACY: sending the lesson corpus to an external provider publishes Outcome lessons externally. Use external generation only after the user deliberately accepts that privacy boundary.
• A non-zero maid enrich validate result is a hard stop. Do not hand-edit .maid/outcomes-digest.json to satisfy validation; regenerate the candidate digest from the bounded corpus instead.
• Use only registered maid enrich and maid insights options. Do not invent provider, model, base URL, timeout, or API-key options for maid-runner.

Workflow

1. Refresh the learned Outcome index when needed:

   maid learn
   

2. Build the bounded corpus the model may see:

   maid enrich prompt --output .maid/outcomes-enrichment-prompt.json
   

   If the index is intentionally stale, use the existing explicit opt-in:

   maid enrich prompt --allow-stale-index --output .maid/outcomes-enrichment-prompt.json
   

3. Send only .maid/outcomes-enrichment-prompt.json to a model.

   Use the model access already available to the hosting agent, or a specific model the user chooses. If that sends the corpus to an external provider, disclose CLOUD-PRIVACY first: sending the lesson corpus externally publishes Outcome lessons outside the local repository boundary.

   The model output must be a candidate .maid/outcomes-digest.json.

4. Validate the candidate digest:

   maid enrich validate --digest .maid/outcomes-digest.json
   

   Treat a non-zero maid enrich validate result as a hard stop. Do not hand-edit .maid/outcomes-digest.json to satisfy validation. Regenerate the candidate digest from the bounded corpus and validate again.

5. Render advisory markdown from the validated digest:

   maid enrich render --digest .maid/outcomes-digest.json --md-output .maid/outcomes-digest.md
   

6. Optionally view deterministic theme aggregation through insights:

   maid insights --theme-map .maid/outcomes-digest.json
   

Output Boundaries

.maid/outcomes-digest.json and .maid/outcomes-digest.md are advisory, generated, ignored artifacts. They are not authoritative Outcome records and must not replace manifest outcome: sections, validation commands, review evidence, or the MAID implementation lifecycle.