The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-supplied input to resolve persona-switching commands and validates JSON creator profiles that may contain untrusted data.
Ingestion points: User request strings processed by scripts/resolve-soul-command.mjs and creator profile data validated by scripts/validate-soul-profile.mjs.
Boundary markers: The files references/mode-contract.md and references/persona-kernel.md provide clear instructions to the agent to treat retrieved content and profiles as data only, explicitly forbidding them from being interpreted as instructions that can override the kernel or expand permissions.
Capability inventory: Execution of local Node.js scripts for command resolution and data validation; no network or high-privilege file system operations are requested outside of reading local reference files.
Sanitization: The skill employs scripts/validate-soul-profile.mjs, which uses regular expression scanning to identify and block forbidden sensitive fields (such as credentials, tokens, or private keys) and absolute local file paths before the data is integrated into the agent context.

activate-tavernweave-soul