The Agent Skills Directory

[COMMAND_EXECUTION]: The script _postplus_shared/00-core/shared-runtime/scripts/download_videos_from_manifest_with_ytdlp.mjs executes the yt-dlp tool via python3 to download video content. This is used to fulfill the skill's purpose of media collection.
[DATA_EXFILTRATION]: The skill manages and reads configuration data, including session tokens, from the PostPlus CLI configuration directory (e.g., ~/.config/postplus/config.json on Linux) to authenticate requests to its hosted collection API.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted data from public Facebook pages, groups, and posts.
Ingestion points: Raw content from Facebook posts, profile titles, and descriptions are collected and stored in .postplus/facebook-research/ (e.g., raw/facebook.json).
Boundary markers: The instructions in SKILL.md do not specify boundary markers or 'ignore' instructions for the agent when processing the results of the research.
Capability inventory: The skill has the capability to execute subprocesses (via yt-dlp), perform network operations to PostPlus APIs, and write files to the local filesystem.
Sanitization: The normalization scripts (internal/public-content/scripts/normalize_public_posts.mjs) extract data fields but do not perform sanitization or filtering to prevent embedded instructions from influencing the agent's behavior during summary generation.

facebook-research