The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-generated notes and conversation history, which represents a potential ingestion surface for untrusted data. Ingestion points: Daily notes, vault notes, and the conversation history as described in SKILL.md and references/promote-mode.md. Boundary markers: The instructions mandate multiple user approval checkpoints ("approval-gated operation", "let the user keep, regroup, or discard", "present candidates before writing") before any durable changes are made. Capability inventory: The skill utilizes file-writing capabilities (save-note, vault safe-write flow) and task-handover tools (project skill). Sanitization: The skill contains an explicit defensive instruction to pass content separately from file paths to avoid shell interpolation risks.- [COMMAND_EXECUTION]: The skill includes proactive security instructions to mitigate command injection risks by requiring the agent to pass content separately from its path rather than using interpolated shell text during the capture process.

ideas