The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local Node.js scripts (resolve-soul-command.mjs and validate-soul-profile.mjs) to perform logic-heavy tasks like command parsing and data validation. These scripts are invoked via the node runtime and operate strictly on local files or command-line arguments provided by the agent.
[DATA_EXPOSURE]: The skill implements proactive security checks in its validate-soul-profile.mjs script. It uses regular expressions to scan for sensitive keywords (e.g., credential, password, secret, private_key) and private Windows absolute paths (e.g., C:\Users\...) within imported profile data to prevent the accidental ingestion or exposure of local secrets.
[INDIRECT_PROMPT_INJECTION]: To protect against malicious instructions embedded in external data, the skill validates 'creator profiles' against a formal JSON schema (profile-schema.json). Furthermore, the persona-kernel.md specifically instructs the agent to treat files and retrieved passages as untrusted inputs that must not be allowed to override the core instructions.
[PERSISTENCE]: The documentation for the 'Portable mode' contract explicitly states that the skill does not write global state files, browser storage, or repository files. This design ensures that persona activations do not create hidden persistence mechanisms across different tasks or sessions.

activate-tavernweave-soul