The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill relies on reading and processing external specification files, which creates a surface for indirect prompt injection attacks.
Ingestion points: The skill reads a spec file from a user-provided path, as well as all source context and decision files referenced by that spec.
Boundary markers: There are no instructions or delimiters defined to separate the specification's data from the agent's operational instructions, nor are there warnings to ignore embedded commands.
Capability inventory: The skill has broad capabilities, including writing and committing source code, creating pull requests, and executing shell commands such as refine and ship as well as project-specific test harnesses.
Sanitization: The instructions do not specify any validation or sanitization of the input specifications before the agent begins executing code changes or tests based on them.

deliver