The Agent Skills Directory

[SAFE]: No malicious patterns or security vulnerabilities were detected during the analysis of the skill's instructions and reference materials.
[DATA_EXPOSURE]: The skill accesses narrative artifacts and writes storytelling seeds to JSON files within a relative project directory. There are no attempts to access sensitive system files (e.g., SSH keys, AWS credentials) or environment variables.
[COMMAND_EXECUTION]: Tool usage is limited to standard Model Context Protocol (MCP) functions such as load_artifact, validate_draft, and log_decision. The skill does not perform arbitrary shell command execution, subprocess spawning, or privilege escalation.
[PROMPT_INJECTION]: The skill instructions define clear boundaries for narrative transformation and genre migration. No patterns of role-play bypass, safety filter overrides, or system prompt extraction were found.
[INDIRECT_PROMPT_INJECTION]: An indirect prompt injection surface exists as the skill processes external 'analysis' artifacts.
Ingestion points: Data enters the workflow through load_artifact("analysis") in SKILL.md.
Boundary markers: The skill enforces output structure via specific JSON schemas (trope_mapping.json, rewrite_seed.json) and validates drafts using the validate_draft tool before writing.
Capability inventory: Capabilities are restricted to writing structured JSON data to project state files and logging decision metadata. No high-privilege tools (network, eval, exec) are present.
Sanitization: Content is mapped to structured fields rather than being passed as free-form prompts, effectively isolating the narrative data from the agent's control flow.

rewrite-trope