The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions require the agent to ingest and analyze data from various local sources to verify user claims, creating a potential surface for indirect prompt injection.\n
Ingestion points: Environment data such as code, git history, configuration files, and documentation as specified in SKILL.md.\n
Boundary markers: The instructions do not define specific delimiters or instructions to the agent to ignore potential commands embedded in the files it reads.\n
Capability inventory: The skill is permitted to use Bash, Read, Grep, and Glob tools to inspect and interact with the local environment.\n
Sanitization: No mechanisms for sanitizing or filtering data retrieved from the environment are described in the instructions.\n- [COMMAND_EXECUTION]: The skill encourages the use of shell commands to gather evidence from the environment, including checking git history and dependency versions.\n
Evidence: SKILL.md instructs the agent to "verify every claim against facts from the environment: code, git history, configs, docs, dependency versions. If you can't find it, go look."

question-it