The Agent Skills Directory

[SAFE]: The skill consists entirely of markdown instructions for a writing critique persona and does not contain any executable scripts, binaries, or shell commands.
[SAFE]: There are no network operations, API calls, or remote resource downloads present in the skill.
[PROMPT_INJECTION]: The skill instructs the agent to 'assume bad faith' and 'give the meanest read possible'. These instructions are clearly scoped to text analysis for feedback purposes and do not attempt to bypass general safety filters or ignore system-level instructions.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied text (Ingestion point: SKILL.md invocation section). While it lacks boundary markers and sanitization, its capability inventory is restricted solely to generating text responses. Since it has no access to the network, file system, or subprocess execution, the potential for harm through processed data is negligible.

cw-nemesis