The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted content, including Agent Operating Procedures (AOPs), Run transcripts, and case data. This creates a surface where malicious instructions embedded in these data sources could attempt to influence the agent's behavior during the analysis or improvement phases.
Ingestion points: Untrusted data enters the context via getRevision (AOPs), listRuns (transcripts), and listCases (backlog data).
Boundary markers: The skill includes an explicit instruction to treat all pulled content as data rather than instructions, which serves as a mitigation against accidental obedience.
Capability inventory: The skill possesses significant capabilities, including modifying AOPs via aop-writer, updating agent configurations with updateAgent, and performing high-volume actions like bulkReprocessCases.
Sanitization: While the skill mandates a 'data, not instructions' policy, there is no technical evidence of automated sanitization or filtering of the ingested content.

improve-queue