The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions consist of creative writing frameworks and worldbuilding methodologies. No markers for bypassing safety filters, disregarding previous instructions, or extracting system prompts were found.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill manages session state by writing creative output to local files (e.g., {world-name}-systemic-{date}.md). It does not contain any network commands (curl, wget) or hardcoded secrets that could lead to data exfiltration.
[OBFUSCATION]: The markdown content is clear and readable. No Base64, zero-width characters, homoglyphs, or other obfuscation techniques were identified.
[REMOTE_CODE_EXECUTION]: The skill does not download or execute remote code. References to other skills (like 'memetic-depth' or 'conlang') are for internal agent routing within the platform's multi-agent framework.
[PRIVILEGE_ESCALATION]: The skill does not attempt to use system-level commands, administrative privileges, or file permission modifications.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill processes user-defined speculative changes (Level 1 Divergence Points) and reads project configuration from context/output-config.md.
Boundary markers: The instructions do not define explicit boundary markers for user data, relying on the generative prompt structure.
Capability inventory: The skill's primary capability is writing text to markdown files in a user-specified directory.
Sanitization: No explicit sanitization of worldbuilding inputs is mentioned, but the risk is minimal given the creative nature of the output.
[DYNAMIC_CONTEXT_INJECTION]: The skill does not utilize the load-time shell execution syntax (!command).

systemic-worldbuilding