peec-learn

SkillMind Learner

Role

Turn project-local Peec outputs into cross-project patterns. Each run does two things:

1. Write — extract 1–3 patterns from a just-produced artifact (decision, brief, zone map, outreach log, learnings.json) and store them in SkillMind with tags so they can be retrieved later.
2. Read — on request, recall patterns matching a project / skill / gap type and hand them back as priors for the next orchestrator cycle.

This is the memory layer beneath peec-report: that skill persists learnings for the project, this skill promotes them across projects.

Input

For write mode:

• project_id — Peec project the artifact came from
• source_skill — which skill produced the artifact (peec-agent, peec-cluster, peec-outreach, peec-content-intel, peec-report)
• artifact_path or artifact_content — the file or inline content to extract from
• optional max_patterns — default 3

For read mode:

• query — what the caller wants to recall (e.g. "editorial outreach DACH high citation rate")
• optional project_id — narrow to patterns originally written for this project
• optional source_skill — narrow to patterns originally written by this skill
• optional k — default 5

Output

Write mode: JSON list of {pattern_id, title, tags, summary} for each persisted pattern, plus a one-line confirmation ("added 3 patterns · skipped 1 dupe").

Read mode: ranked list of {pattern_id, title, summary, provenance: {project_id, source_skill, date}, score}. Empty list is a valid result — say so plainly.

Neither mode produces dashboards.

When to use

Write:

• Right after peec-report emits learnings.json
• After a peec-outreach batch closes (week-end ritual)
• After peec-cluster ships a zone map (zones become reusable taxonomy patterns)
• After peec-agent logs a decision whose 4-week metric came in (attribution is known)
• After peec-content-intel ships a brief that later won its prompt (write the retrospective pattern, not the brief itself)