lore
Pass
Audited by Gen Agent Trust Hub on Sep 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill represents an attack surface for indirect prompt injection because it is designed to ingest and synthesize data from multiple untrusted sources.
- Ingestion points: The skill scans agent journals (stored in
.agents/*.md), Triage postmortems, and Mend remediation logs. These files are populated by other agents or external processes whose outputs may be influenced by adversarial input. - Boundary markers: While the skill requires reading full source entries and performing semantic clustering, it lacks explicit instructions to identify or neutralize embedded instructions within the journal text that might attempt to manipulate the cataloging process.
- Capability inventory: The skill has the capability to read a wide variety of files across the agent ecosystem and write consolidated insights into
METAPATTERNS.mdand.agents/lore.md. - Sanitization: There is no evidence of content sanitization, escaping, or filtering for natural language instructions that might be embedded in the data being processed. A malicious entry in a journal could potentially influence the 'confidence' or 'freshness' of patterns recorded in the institutional memory.
Audit Metadata