The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes functionality to ingest and analyze external, untrusted codebases through scripts/scan_patterns.py.
Ingestion points: scripts/scan_patterns.py reads the content of arbitrary text files within a target directory provided as a command-line argument.
Boundary markers: While the skill documentation in references/implementation-patterns.md strongly advocates for the use of boundary markers (e.g., <untrusted_user_data> tags), the heuristic scanner script itself reads raw file content directly into regex engines.
Capability inventory: The skill has the capability to read local files, write JSON results to stdout, and execute internal scripts via subprocess.run in the test suite.
Sanitization: The scanner does not sanitize the input files; it performs heuristic regex matching to identify potential vulnerabilities for human review.
[COMMAND_EXECUTION]: The testing script scripts/test_skill.py utilizes the subprocess.run module to execute other scripts within the skill's package.
Evidence: scripts/test_skill.py executes scripts/validate.py and scripts/scan_patterns.py using sys.executable.
Context: These executions are used for validating the skill's integrity and functionality. The commands are constructed using fixed paths relative to the skill root, minimizing injection risks.

secure-ai-agent-coding