The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied input via the $ARGUMENTS variable, which is then interpolated into a call to another skill. \n- Ingestion points: $ARGUMENTS in SKILL.md. \n- Boundary markers: None present. \n- Capability inventory: Invokes the Skill tool and allows Bash tool execution restricted to scripts/larch.sh. \n- Sanitization: The skill provides explicit instructions to the agent to filter out restricted flags and enforces a strict digits-only format for positional arguments, effectively mitigating basic injection risks. \n- [SAFE]: The skill's instructions are consistent with its stated purpose as a functional shortcut. No patterns associated with obfuscation, unauthorized persistence, credential harvesting, or remote code execution from unknown sources were detected.

im