The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill aggregates and synthesizes information from various agent journals and log files, creating a potential surface for indirect prompt injection if those sources contain untrusted external data. The risk is effectively mitigated by the skill's structured processing pipeline and governance model.
Ingestion points: Consumes data from .agents/*.md journals, Triage postmortems, and Mend remediation logs.
Boundary markers: Uses a multi-stage synthesis lifecycle (Formation, Evolution, Retrieval) and bi-temporal validity tracking (observation time vs. ingestion time) to isolate and validate information.
Capability inventory: Capable of writing to the METAPATTERNS.md catalog and broadcasting LORE_INSIGHT or LORE_ALERT signals to other agents like Architect and Darwin.
Sanitization: Implements strict semantic similarity clustering thresholds (≥ 80%) and requires human-in-the-loop review for knowledge graph mutations.

lore