The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs the user to install an external CLI tool (sem-cli) via Homebrew or by direct binary download from an unverified GitHub repository (Ataraxy-Labs/sem) as seen in the CI workflow configuration.
[COMMAND_EXECUTION]: The skill performs broad file system operations, including copying a shell script (pre-commit-hook.sh) into the .git/hooks/ directory and using chmod +x to make it executable. This script is then configured to run automatically during every git commit process, establishing a persistence mechanism on the user's local environment.
[REMOTE_CODE_EXECUTION]: The provided GitHub Action workflow (github-action-guard.yml) implements a pattern that downloads a binary from a remote URL, extracts it to a system path (/usr/local/bin), and sets executable permissions. This constitutes execution of unverifiable code from an external source.
[PROMPT_INJECTION]: The skill contains a vulnerability surface for indirect prompt injection by processing untrusted codebase files to generate instructions for AI agents.
Ingestion points: Files such as package.json, go.mod, and requirements.txt are read and their contents are used to fill placeholders in the AGENTS.md template.
Boundary markers: The skill does not implement boundary markers or instructions to ignore embedded directives when interpolating external content into the generated documentation.
Capability inventory: The skill has access to file-write capabilities and the ability to execute subprocesses, which could be exploited if malicious content is injected via the codebase files.
Sanitization: There is no evidence of sanitization or validation of the data extracted from the project files before it is incorporated into the agent's operating instructions.

cali-agents-md-generator