The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which could contain malicious instructions designed to manipulate the agent.
Ingestion points: The skill ingests data from web pages via context_dev_scrape_markdown, YouTube captions via youtube_web_v2_video_captions (as defined in references/sandbase-api-map.md), and user-provided meeting transcripts or raw notes.
Boundary markers: The instructions do not define specific delimiters or provide guidance to the agent to disregard embedded instructions within the source material.
Capability inventory: The skill can invoke internal tools using sandbase_call_tool and has instructions for publishing results to external task trackers like GitHub or Jira if authorized.
Sanitization: There is no mention of filtering or sanitizing the input data before processing it for summary generation or action item creation.
[DYNAMIC_EXECUTION]: The skill uses a dynamic resolution pattern to invoke capabilities.
Pattern: It resolves tool schemas at runtime using sandbase_describe_tool before execution via sandbase_call_tool.
Context: This resolution is guided by a static mapping file (references/sandbase-api-map.md) and appears to be standard vendor functionality for the sandbaseai platform.
[COMMAND_EXECUTION]: The skill is configured to execute specific data retrieval tools.
Evidence: References the use of context_dev_scrape_markdown and youtube_web_v2_video_captions to fulfill its purpose of gathering meeting source material.

meeting-minutes