The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute various shell commands to perform the audit, including running a test harness (python tests/run_tests.py), directory listing (ls -la), file comparison (diff -q), and pattern searching (grep). These commands are used to verify the integrity of the local system state.
[DYNAMIC_EXECUTION]: The skill utilizes a Python one-liner (python -c "import sys; sys.path.insert(0,'tests'); ...") to dynamically resolve the SKILLS_DIR at runtime. It also executes a local synchronization script (python scripts/check_desktop_sync.py) to verify mirror status. These operations are standard for a maintenance tool but involve executing code determined at runtime.
[DATA_EXPOSURE]: The audit process involves reading sensitive local configuration files, specifically ~/.claude/CLAUDE.md and the contents of ~/.claude/agents/. This access is necessary for identifying configuration drift between the local environment and the project's registry, and the information is not exfiltrated.
[INDIRECT_PROMPT_INJECTION]: The skill has a large attack surface for indirect prompt injection because it reads and parses numerous markdown files, logs, and reports (e.g., weekly-health-check-*.md, agent_registry.md, overlays).
Ingestion points: Reads multiple project files including reports, registry manifests, and individual agent skill files.
Boundary markers: None explicitly defined in the instructions to separate data from instructions during parsing.
Capability inventory: The skill has access to shell execution (python, grep, diff, ls) and sensitive file reading.
Sanitization: There is no evidence of sanitization or escaping of the content read from these files before it is processed by the agent.

system-audit