The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes a workflow for fetching and processing external content from the web while having access to powerful tools such as Bash and file system operations. This provides an attack surface where malicious instructions hidden in external web pages could potentially influence the agent's subsequent actions.
Ingestion points: The skill retrieves data from external websites and documentation using the WebSearch and WebFetch tools as described in the "Modes" section of SKILL.md.
Boundary markers: There are no explicit instructions for the agent to use XML-like tags, delimiters, or specific system instructions to treat the fetched content purely as data, which would help prevent the agent from obeying instructions embedded in that data.
Capability inventory: The skill has access to Bash, Read, Write, Grep, and Glob (listed in the allowed-tools section of SKILL.md).
Sanitization: The skill does not define any validation or sanitization steps for the data retrieved from external sources before it is analyzed or used to generate technical documentation.

research