The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses URL encoding (specifically %72 for the character 'r') to mask the path to scripts/lint_readme.py and references/readme-principles.md in its primary instructions and reference files. This is a form of obfuscation that can be used to bypass simple text-based security filters.
[EXTERNAL_DOWNLOADS]: The scripts/lint_readme.py script downloads a configuration package for the Vale linter from github.com/vale-cli/Google. This is a well-known source for linting rules and is used to initialize the linter environment.
[COMMAND_EXECUTION]: The skill executes external processes via subprocess.run to perform repository inspection and linting. The implementation follows security best practices by using argument lists rather than shell strings, reducing the risk of command injection.
[DATA_EXFILTRATION]: The skill contains an attack surface for indirect prompt injection via scripts/inspect_project.py.
Ingestion points: It recursively reads files matching TEXT_SUFFIXES (including .md, .json, .yaml) and project manifests from the repository root.
Boundary markers: The instructions do not require the agent to use specific delimiters when reading these files, although it advises prioritizing manifests over prose.
Capability inventory: The skill can execute the Vale linter and read arbitrary text files within the repository.
Sanitization: No content sanitization or escaping is performed on the repository data before it is presented to the agent context.

awesome-readme