The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is instructed to execute a local shell script at hack/spec_metadata.sh to generate metadata for documentation. It also leverages system tools including git and gh (GitHub CLI) to retrieve branch, commit, and repository information.
[EXTERNAL_DOWNLOADS]: If requested by the user, the skill invokes a web-search-researcher sub-agent to fetch data from external websites and resources.
[PROMPT_INJECTION]: The skill processes untrusted data from the codebase and external web searches, creating an indirect prompt injection vulnerability.
Ingestion points: The skill reads user-specified files in their entirety and processes results from web searches.
Boundary markers: There are no explicit delimiters or instructions provided to the agent or its sub-agents to ignore or isolate instructions that may be embedded within the files or web content being researched.
Capability inventory: The skill has the ability to read arbitrary files, create directories, write documentation to the filesystem, and execute local scripts.
Sanitization: The skill lacks any description of sanitization or validation logic to filter content retrieved from the codebase or the internet before it is synthesized into research reports.

research