The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to 'mine' data from potentially untrusted sources which could contain malicious instructions.
Ingestion points: The skill reads git history via git log, performs searches across the codebase using rg (specifically targeting comments like TODO, HACK, and FIXME), and fetches data from GitHub issues and pull requests via the gh CLI.
Boundary markers: The instructions do not define delimiters or specific 'ignore embedded instructions' prompts when processing the mined text.
Capability inventory: The skill possesses the ability to write to several critical project files, including README.md, CONTRIBUTING.md, and internal skill definitions in skills/internal/project-context/.
Sanitization: There is no evidence of sanitization or filtering of the content retrieved from external or community-contributed sources (like PR comments or issues) before it is proposed as project context.
[COMMAND_EXECUTION]: The skill logic relies on the execution of multiple shell commands to gather project metadata.
Evidence: The skill instructions specify the use of git remote -v, git log, rg (ripgrep), and gh (GitHub CLI). While these are standard developer tools, their automated execution on a repository represents a capability surface that should be monitored.

setup-project-context