The Agent Skills Directory

[COMMAND_EXECUTION]: The diagnostic scripts execute local binaries such as claude, codex, and ripgrep (rg) to gather version data and search log files. These executions are constrained to specific diagnostic purposes and use controlled argument patterns.
[EXTERNAL_DOWNLOADS]: The check_updates function in scripts/agent_health.py queries downloads.claude.ai to verify version freshness. This targets a well-known service associated with the agent's developers.
[DATA_EXFILTRATION]: The skill accesses sensitive configuration files, including ~/.claude.json and settings.json, and scans session transcript logs (*.jsonl). While intended for local auditing, the contents are loaded into the agent's context. The skill's instructions specifically warn against printing secrets found during this process.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its ingestion of external session logs.
Ingestion points: Session logs (~/.claude/projects/*.jsonl, ~/.codex/sessions/**/*.jsonl) and local JSON/TOML configuration files.
Boundary markers: The skill employs structured JSON and TOML parsing logic that identifies and reports malformed records rather than processing them as instructions.
Capability inventory: The skill has the ability to execute shell commands via subprocess.run to inspect system tools and search files.
Sanitization: Log data is parsed as structured objects, and command-line interactions are handled via shlex to minimize injection risks.

skill-usage-stats