The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes bundled local scripts (scripts/memory-scan.py and scripts/cache-report.sh) and common utilities like jq, grep, and env to extract system configuration and usage metrics.
[CREDENTIALS_UNSAFE]: The skill inspects environment variables containing sensitive information, such as ANTHROPIC_API_KEY and ANTHROPIC_AUTH_TOKEN, and reads session logs that contain conversation history. This data is used solely to audit and report on token usage patterns.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources including project-specific CLAUDE.md files and session logs. This represents a potential surface for indirect prompt injection if those files contain adversarial instructions.
Ingestion points: CLAUDE.md, CLAUDE.local.md, and project session logs (.jsonl).
Boundary markers: No explicit boundary markers are used when interpolating file content into the audit report.
Capability inventory: The skill has access to shell execution and file reading capabilities to perform its analysis.
Sanitization: Content from files is processed to calculate sizes and tokens without additional sanitization or escaping.

token-audit