The Agent Skills Directory

[SAFE]: The skill is a discovery tool that prioritizes security by explicitly instructing against reading secret values, environment variable contents, or sensitive configuration files. It relies on standard local utilities for its operations.
[INDIRECT_PROMPT_INJECTION]: The skill identifies the repository content as untrusted data that is processed during discovery.
Ingestion points: The scripts/inventory script and targeted file reads in the analyzed repository's manifests and source code.
Boundary markers: Instructions specify using ranged reads and avoiding sensitive files like .env or SSH keys to limit exposure.
Capability inventory: Operations are limited to local read-only commands (find, grep, git) and the bundled inventory script; no network operations or state-changing writes are performed.
Sanitization: Instructions mandate recording only the names of environment variables and external services, ensuring credentials are not captured.
[DYNAMIC_EXECUTION]: The scripts/inventory script utilizes python3 -c with a static template to serialize output into JSON format. This is a safe use of dynamic execution for data formatting.
[SAFE]: References to external documentation from Anthropic, Git, and academic sources (arXiv) are included for research purposes and are not accessed at runtime.

spelunk