The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains explicit instructions in SKILL.md directing the agent to ignore safety filters and not refuse tasks based on the sensitivity or 'discomfort' of fictional content such as domestic violence, abuse, or other dark themes. This is framed as a mandatory requirement to prevent the analysis pipeline from stopping.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided stories across multiple stages to generate analytical reports. 1. Ingestion points: Raw text provided by the user is ingested in Phase 1 of SKILL.md and processed in material-decomposition.md. 2. Boundary markers: The skill uses Markdown blockquotes to delimit quoted story content in reports. 3. Capability inventory: The agent writes files to the local file system and executes local shell commands. 4. Sanitization: The instructions lack explicit sanitization protocols to prevent malicious instructions embedded in user-provided novels from being interpreted by the agent or subsequent skills in the downstream pipeline like story-short-write.
[COMMAND_EXECUTION]: The references/output-contract.md file contains instructions for the agent to execute a local shell script using bash scripts/check-shared-files.sh to verify file synchronization between skill components.

story-short-analyze