The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided narrative text and story descriptions to diagnose "story states." This processing of untrusted data represents an indirect prompt injection surface. The skill instructions lack explicit boundary markers or sanitization requirements to isolate potentially malicious instructions embedded within the user's story content from the agent's operational logic.
Ingestion points: User-provided story descriptions and narrative problems processed by the agent (defined in SKILL.md).
Boundary markers: Absent; the instructions do not specify delimiters for user-provided narrative content.
Capability inventory: File-writing for diagnostic persistence (e.g., to explorations/story-diagnostics/) and local script execution via Deno.
Sanitization: Absent; the skill does not require escaping or filtering of external narrative content before processing.
[COMMAND_EXECUTION]: The skill utilizes Deno to run local helper scripts (entropy.ts and functions.ts) for generating creative elements. These scripts are executed with the --allow-read flag to access local JSON data files, establishing a pattern of local code execution within the agent's environment to support the skill's diagnostic functions.

story-sense