The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local file system, such as system metrics, logs, and directory structures, which creates a surface for indirect prompt injection attacks.
Ingestion points: The scripts scripts/analyze_throughput.py and scripts/identify_bottlenecks.py recursively read and analyze file system content and directory structures based on user-provided paths.
Boundary markers: The instructions in SKILL.md lack explicit delimiters or instructions for the agent to ignore potentially malicious instructions embedded within the analyzed performance data or file contents.
Capability inventory: The skill is configured with access to powerful tools including Bash, Read, Write, and Grep, which could be abused if an injection is successful.
Sanitization: There is no evidence of content validation, escaping, or sanitization of the ingested file data before it is incorporated into the agent's context for analysis.

analyzing-system-throughput