The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests benchmark manifests and research artifacts to create persona definitions and visual consistency packs.
Ingestion points: Research data and benchmark manifests are ingested from external JSON files (e.g., manifest.json, input.json) via readJson calls in scripts/build_persona_pack.mjs and _postplus_shared/00-core/shared-runtime/scripts/download_videos_from_manifest_with_ytdlp.mjs.
Boundary markers: No explicit boundary markers or instructions to disregard embedded commands within the ingested data were found in the persona generation logic.
Capability inventory: The skill possesses capabilities for subprocess execution (yt_dlp), network operations through a custom runtime, and file system write access.
Sanitization: The skill lacks explicit sanitization or filtering of the benchmark data before using it to generate persona locks and image prompt packs.
[COMMAND_EXECUTION]: The skill executes external system commands to process video content.
Evidence: The script _postplus_shared/00-core/shared-runtime/scripts/download_videos_from_manifest_with_ytdlp.mjs utilizes the node:child_process module to spawn python3 -m yt_dlp for downloading videos from URLs specified in the manifest.
[DATA_EXFILTRATION]: The skill is designed to access sensitive local configuration files to manage session authentication with the vendor's API.
Evidence: _postplus_shared/00-core/shared-runtime/scripts/lib/postplus_cli_config.mjs reads authentication tokens (cliSessionToken) from the user's local PostPlus configuration directory (e.g., ~/.config/postplus/config.json). These tokens are used to authorize requests to the vendor's hosted capability bridge.

persona-pack