The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to treat the user-editable todo list as its 'source of truth', which could allow a user to inject instructions via task titles or notes. * Ingestion points: User-provided content enters the agent's context through the shared todo list tool which reflects UI-based edits. * Boundary markers: No instructions are provided to help the agent distinguish between valid task descriptions and potentially malicious instructions embedded in the user-provided data. * Capability inventory: The agent possesses capabilities to update plan statuses, generate task summaries, and provide commit messages for the host system. * Sanitization: There is no mention of validating or sanitizing user-provided content before the agent processes it.

todos