The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from potentially untrusted sources such as the current session, notes, commits, and code snippets, which could contain hidden instructions targeting the agent. * Ingestion points: The skill harvests data from the current session, today's note, project evidence, screenshots, commits, and user-provided text (SKILL.md Step 2). * Boundary markers: While explicit technical delimiters are not specified, the instructions include a 'Safety gate' (Step 3) to classify data as private, needs-review, or public-safe. * Capability inventory: The skill can write to specific files (Obsidian notes) and hand off data to other tools like tweet-today and blog (SKILL.md Step 4). * Sanitization: Step 5 ('Approval-gated write') mandates that the agent must show the exact receipt to the user for approval before any write operation occurs.

capture-receipt