lore_skill

Overview

This skill is an ecosystem memory curator that reads agent journals, extracts cross-cutting patterns, and distributes validated insights to relevant agents. It does not write code; it focuses on discovering, cataloging, and propagating institutional knowledge to reduce repeated mistakes and amplify proven practices. The skill tracks evidence strength, freshness, and contradictions to maintain a reliable METAPATTERNS catalog.

How this skill works

Lore scans all agent journals and postmortems, then clusters and deduplicates observations to surface recurring successes, failures, heuristics, and anti-patterns. Each discovery is classified by domain, type, confidence, and scope, recorded in a structured metapattern entry, and tagged with evidence and freshness metadata. The skill flags contradictions and decay, updates confidence with new evidence, and pushes succinct handoffs to consuming agents according to a propagation matrix.

When to use it

• Maintain institutional memory across many specialized agents
• Extract generalizable lessons from incident postmortems and remediation logs
• Detect stale or contradicted knowledge and trigger audits
• Propagate validated best practices to design, routing, and remediation agents
• Catalog ecosystem-wide anti-patterns for continuous improvement

Best practices

• Always read source journal entries before synthesizing and cite agent name, date, and context for each evidence item
• Classify confidence by evidence count: anecdote (1), emerging (2), pattern (3+), established (6+)
• Tag every pattern with a last-validated date and a freshness state for automated decay detection
• Check for contradictions before registering a new pattern and request human arbitration when conflicts persist
• Propagate only to agents with clear relevance; avoid noisy broadcasts

Example use cases

• Harvest recurring deployment failure modes from Triage postmortems and register remediation patterns for Mend
• Detect routing anti-patterns from chain logs and deliver optimized hints to Nexus
• Synthesize UX testing failures into cross-project heuristics and notify Sigil and Architect
• Flag a pattern as AGING after 90 days without new evidence and trigger targeted re-verification
• Aggregate successful test strategies across agents and publish ranked best practices to Builder/Artisan