The Agent Skills Directory

[COMMAND_EXECUTION]: The instructions explicitly direct the agent to run a shell command: uv run --with lxml --with pyyaml python3 ~/projects/workflows/scripts/prose-audit.py. This execution of a script from a hardcoded local path (~/projects/) is a high-risk pattern for command injection and arbitrary execution.
[DYNAMIC_EXECUTION]: The use of uv run involves the dynamic installation of Python packages (lxml, pyyaml) and the runtime execution of code, which can be exploited if the source script or package registry is compromised.
[INDIRECT_PROMPT_INJECTION]: The skill provides a mechanism to process untrusted text drafts through a command-line utility.
Ingestion points: The text draft is passed directly as a command-line argument to the audit script in SKILL.md.
Boundary markers: The skill lacks explicit instructions to sanitize or delimit the draft content to prevent it from being interpreted as commands by the underlying script.
Capability inventory: The agent is granted the ability to execute arbitrary shell commands and interact with the local file system.
Sanitization: No evidence of input validation or escaping for the draft content is present in the skill instructions.

writing-general