The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill accepts input via the $ARGUMENTS variable and forwards it verbatim to the implement skill, establishing a surface for indirect prompt injection vulnerabilities.
Ingestion points: The $ARGUMENTS placeholder in the Usage and Behavior sections in SKILL.md.
Boundary markers: Absent. There are no delimiters or instructions provided to the agent to treat the forwarded arguments as untrusted data.
Capability inventory: The skill utilizes the Skill tool to execute other components and is authorized to use Bash (scoped to scripts/larch.sh) and Read tools.
Sanitization: Absent. The skill performs no filtering, validation, or escaping of the forwarded arguments before execution.

f