The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external content fetched from LinkedIn, which is an untrusted source and presents a surface for indirect prompt injection attacks.
Ingestion points: The skill reads post bodies and comments through the Apify MCP connector or Handlus browser automation as described in SKILL.md (Step 2).
Boundary markers: SKILL.md contains a robust 'Untrusted content' section that explicitly defines how to handle fetched data and warns the agent to ignore any embedded instructions.
Capability inventory: The agent has network access to fetch data but no capabilities for file-write, subprocess execution, or direct command execution are present in the provided scripts.
Sanitization: The skill implements a humanizer pass and instruction-level filtering to ensure that drafted comments are not influenced by adversarial content within the fetched posts.

linkedin-comment-drafter