The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a local Node.js script (render-methodology-skill.mjs) via the shell to transform input data into a new skill file. The execution is bounded to local paths and internal logic.
[DATA_EXFILTRATION]: The underlying scripts perform local file system reads using fs.readFileSync to access methodology manifests and definitions. Path traversal is mitigated by validating input identifiers against a known manifest and using strict regular expressions for file names.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from a JSON input (e.g., task, context, skill_summary) and interpolates these values into a generated SKILL.md. This creates a surface where malicious instructions embedded in the input could be carried over into the output skill.
Ingestion points: Data from input.json is processed by the core rendering logic in scripts/methodology-skill-core.mjs.
Boundary markers: The skill contains explicit 'Boundaries' and 'Stop' sections intended to limit the agent's behavior during the creation process.
Capability inventory: The skill uses fs.readFileSync for file access, fs.existsSync for directory checks, and node command execution for rendering. The agent is ultimately tasked with writing the output to the repository.
Sanitization: The implementation uses extensive validation, including regex checks for skill names, character count limits, and verifying methodology references against a local manifest file.

methodology-skill-creator