The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data retrieved from Facebook pages, groups, and comments. This creates a surface for indirect prompt injection where malicious instructions embedded in the source content could influence the agent's behavior during analysis. 1. Ingestion points: Data is ingested via various postplus research run commands (e.g., facebook-comments, facebook-group-posts) as defined in SKILL.md and references/public-content.md. 2. Boundary markers: No specific boundary markers or instructions to ignore embedded commands were found in the processing logic. 3. Capability inventory: The agent can execute CLI commands (postplus research run) and write to the local file system (result.json, evidence.html). 4. Sanitization: There is no explicit sanitization or filtering of the retrieved content before it is passed to the LLM for analysis or written to artifacts.
[COMMAND_EXECUTION]: The skill relies on the execution of the postplus CLI tool for all data collection and reporting tasks. While these commands are central to the skill's primary research purpose, they involve interaction with external services and local file writing. Evidence: Multiple references to postplus research run, postplus ad-library report, and postplus doctor throughout the skill files.

facebook-research