The Agent Skills Directory

[DATA_EXFILTRATION]: The skill contains instructions to search for and write to the ~/.claude/projects/ directory. This path is used by the agent platform to store project-specific metadata, session information, and configuration. Accessing this directory allows the skill to interact with the agent's internal state, potentially exposing project information or session history.- [COMMAND_EXECUTION]: In the SELF-HEALING VALIDATION section, the skill is instructed to run the project's test suite and build/compile commands. This allows the skill to execute arbitrary code defined in the build scripts and tests of the project being audited.- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes user-controlled $ARGUMENTS to guide audit phases without sanitization or boundary markers. 1. Ingestion points: User-provided $ARGUMENTS defining the system or modules to audit in SKILL.md. 2. Boundary markers: Absent; no delimiters are used to wrap external input. 3. Capability inventory: Subprocess calls for running build/compile and test suites in SKILL.md. 4. Sanitization: Absent; the input is used directly to drive the audit workflow.

healthcare-audit