The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes specifications and task tickets which are external inputs, creating a vulnerability to indirect instructions.
Ingestion points: Technical specifications and implementation tickets are ingested in the first step of the process in SKILL.md.
Boundary markers: The skill lacks explicit delimiters or instructions to the agent to treat input data purely as data and not as command overrides.
Capability inventory: The skill possesses significant capabilities across its workflow, including subagent creation, writing files to locations outside the repository, and performing complex git operations such as branch merging and worktree management.
Sanitization: No data validation or sanitization process is described for the incoming specification content before it is passed to implementation subagents.

implement-spec