The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from project files, including READMEs, manifests, and installer scripts. This creates a surface for indirect prompt injection where malicious instructions embedded in a project's documentation could attempt to influence the agent's behavior.
Ingestion points: The skill reads various project files using Read, Glob, and Grep tools (e.g., package.json, installer scripts, README.md).
Boundary markers: The skill instructions include a strong security invariant (Invariant 2) that explicitly forbids the agent from quoting file content verbatim, requiring it to paraphrase findings in its own words.
Capability inventory: The skill possesses Edit and Write capabilities for applying fixes, and uses read-only Bash commands for Git operations.
Sanitization: The skill implements a content-signature filter to detect and exclude files containing credentials or high-entropy tokens before they are processed.
[EXTERNAL_DOWNLOADS]: The documentation references installation from the author's official GitHub repository and marketplace.
Evidence: The README provides installation instructions via https://github.com/pilniczek/dev-skills and https://skills.sh/pilniczek/dev-skills.
Note: These sources are associated with the skill's author ('pilniczek') and represent official vendor resources.
[COMMAND_EXECUTION]: The skill uses specific shell commands to interact with the project's version control system.
Evidence: allowed-tools includes Bash(git status:*), Bash(git diff:*), and Bash(git log:*).
Context: These are read-only Git operations used to identify recently changed files to improve audit accuracy.

docs-consistency-check