The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided research files (venue info, player preferences) located in /mnt/agents/output/research/. These files are passed to subagents as context without sanitization. If these files contain adversarial instructions, they could influence the subagents' behavior. • Ingestion points: User-uploaded files in /mnt/agents/output/research/ are referenced in all subagent task prompts. • Boundary markers: No delimiters or ignore instructions are specified for the interpolation of user data. • Capability inventory: The skill can create subagents, dispatch tasks, write files to the /mnt/agents/output/ directory, and execute local Python scripts. • Sanitization: No explicit validation, escaping, or filtering of external content is mentioned.
[COMMAND_EXECUTION]: The orchestrator is instructed to execute a local Python script to verify word counts of generated character scripts and host guides. • Evidence: 'python skills/general-writing/scripts/check_wordcount.py --min {target} --lang auto' in SKILL.md. This is a local script execution pattern.

murder-mystery