The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local system utilities to process input files. Specifically, it uses the macOS sips tool for converting HEIC images to JPEG and the cp command for managing temporary files in the local environment during processing.
[EXTERNAL_DOWNLOADS]: The skill's installation process involves fetching resources from a vendor-owned repository located at github.com/peter-tu-zynkr/zynkr-skill-builder. These resources are consistent with the skill's stated purpose and author context.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from external transcripts (e.g., Fireflies recaps) and whiteboard images. It mitigates potential injection risks through a structured evidence chain:
Ingestion points: Transcript data (Google Docs, Gmail content, or pasted text) and whiteboard photos (Google Drive or local files) are ingested in Step 1.
Boundary markers: The skill enforces strict tab schemas, MECE (Mutually Exclusive, Collectively Exhaustive) classification rules, and a mandatory verbatim pass (② 白板原文) that separates raw data from processed project items.
Capability inventory: Capabilities include Google Workspace operations (read_sheet_values, modify_sheet_values, create_sheet, get_doc_as_markdown, create_spreadsheet, update_drive_file, mcp__google-workspace__draft_gmail_message) and local file/image operations (sips, cp, PIL).
Sanitization: Handwriting readings with low confidence are flagged for user review in the README tab rather than being silently adopted. The skill also performs coverage checks against existing plan documents to ensure data consistency.

planning-session-synth