The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface due to its instruction to ingest untrusted data and follow directives found within that data.
Ingestion points: As part of its core workflow, the skill reads and analyzes the repository of the project it is auditing, which is untrusted external data.
Boundary markers: Absent. The skill lacks instructions or delimiters to isolate the codebase data from the agent's internal logic. In contrast, the 'Overrides' section in SKILL.md commands the agent to 'Pay attention to specific rules and instructions in the project's documentation and prompt files which may require you to override certain best practices.'
Capability inventory: The agent possesses the capability to read files, generate a markdown report (security_best_practices_report.md), and perform code modifications and git commits as specified in the '# Fixes' section of SKILL.md.
Sanitization: Absent. There is no requirement for the agent to sanitize or ignore instructions embedded in the audited files, nor are there validation steps to prevent malicious content from influencing the generated report or code fixes.
[PROMPT_INJECTION]: The '# Overrides' section in SKILL.md includes the directive 'do not fight with them' regarding user-requested bypasses of security best practices. While intended for user flexibility, this instruction weakens the agent's resistance to adversarial content within the codebase that may attempt to deceptively justify the removal of security controls.

security-best-practices