The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted content from local or remote repositories to generate documentation.
Ingestion points: Reads project metadata (pyproject.toml, package.json), source code (main.py, cli.py), and existing documentation (README.md, CONTRIBUTING.md) in SKILL.md Step 2.
Boundary markers: None identified. There are no instructions for the agent to ignore potentially malicious embedded instructions in the scanned codebase.
Capability inventory: Executes shell commands (gh repo clone, git log, git tag) and writes files to the project directory or /tmp/ in SKILL.md Step 1, 2, and 7.
Sanitization: No explicit sanitization or filtering of external content is mentioned before it is processed by the LLM for README generation.
[COMMAND_EXECUTION]: The skill uses shell commands to interact with Git and GitHub.
Evidence: SKILL.md uses gh repo clone owner/repo /tmp/readme-target, git log --oneline -30, and git tag --sort=-v:refname | head -5 to gather context. While these are standard development workflows, executing them on untrusted remote repositories involves inherent risk.

readme-generation