The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructions direct the agent to scan sensitive shell profile and environment files (including ~/.zshrc, ~/.zshenv, ~/.zprofile, ~/.profile, ~/.bashrc, ~/.bash_profile, and .env*) for exported credentials. While this is part of a security best practice to avoid users pasting secrets into the chat, reading these files exposes local system configuration data to the agent context.\n- [COMMAND_EXECUTION]: The skill uses the Bash tool to perform grep operations on local configuration files and to execute curl commands for probing user-provided backend URLs to verify connectivity.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input which is then used to influence command execution and downstream skill handoffs.\n
Ingestion points: Search queries and environment variable names provided by the user in SKILL.md.\n
Boundary markers: Absent; the skill does not explicitly instruct the agent to use delimiters or ignore instructions when passing user data to the /search skill.\n
Capability inventory: Accesses the file system and network via the Bash tool and can trigger other skills via the Skill tool.\n
Sanitization: Implements an "indirect credentials" pattern by requiring variable names rather than raw secrets, which prevents sensitive tokens from being recorded in the prompt history.

quickstart