The Agent Skills Directory

[SAFE]: The skill operates exclusively on local agent journals and project logs within the repository to synthesize institutional knowledge. It does not utilize any high-risk tools such as network requests, administrative command execution, or persistence mechanisms. Its functions are confined to reading text data and updating a central markdown-based catalog (METAPATTERNS.md) based on predefined evidence thresholds.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests untrusted data from various agent journals (.agents/*.md) and postmortem reports. However, the risk is mitigated by the skill's 'Core Contract' which requires multi-source evidence (confidence scoring), bi-temporal validity tracking, and mandatory human review for any architectural or structural changes to the knowledge graph. These structural safeguards align with best practices for managing autonomous agent memory.
[DATA_EXPOSURE]: While the skill accesses potentially sensitive information such as incident remediation logs and postmortems, this access is required for its primary purpose of extracting failure patterns. The skill does not contain any exfiltration vectors, and its output is limited to propagating high-confidence insights to other authorized agents in the ecosystem.

lore