The Agent Skills Directory

[NO_CODE]: The entire skill is comprised of markdown documentation, instructional phases, and prompt templates. There are no executable files (e.g., .py, .js, .sh) or scripts included in the package.
[SAFE]: The skill operates by saving and reading session logs in a dedicated local directory (.omc/distill-sessions/). It does not attempt to access sensitive system files (such as .ssh, .aws, or .env), nor does it include commands for network exfiltration like curl or wget.
[PROMPT_INJECTION]: The skill facilitates the analysis of external data (previous model outputs) which constitutes an indirect prompt injection surface.
Ingestion points: Untrusted data enters the context when the user or agent pastes model outputs into files within .omc/distill-sessions/ for analysis in Phase 1 and Phase 2.
Boundary markers: The instructions use structured markdown headers and explicit XML-style tags (<extraction>) to separate the data being analyzed from the reasoning instructions.
Capability inventory: The skill itself provides no automation or executable capabilities (e.g., no subprocess calls, network access, or eval functions), relying entirely on the host agent's native reasoning abilities.
Sanitization: No specific sanitization or filtering is prescribed for the ingested data, but the lack of executable capabilities mitigates the risk of instructions in the data being inadvertently executed.

distill-session