The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, potentially untrusted content from meeting transcripts and AI summaries which could contain malicious instructions.
Ingestion points: Meeting transcripts, summaries, and action items are retrieved via the dws minutes command suite, specifically +transcript and +detail in SKILL.md and references/minutes.md.
Boundary markers: The instructions mandate structured data retrieval using the --format json flag and require explicit user confirmation for all write operations, as detailed in the '最小 DWS 执行契约' section of SKILL.md.
Capability inventory: The skill has the ability to write to DingTalk (updating titles, sharing permissions), execute the dws command-line tool, and perform local file writes (via +export-pack and the minutes_recent_summary.py script).
Sanitization: The skill relies on the structured nature of the CLI output (JSON) and the agent's safety guardrails, as no explicit text sanitization is performed within the provided Python scripts.
[COMMAND_EXECUTION]: The script scripts/minutes_recent_summary.py executes the dws utility using subprocess.run. The implementation is safe as it avoids shell=True and strictly defines the command list, starting with the trusted dws binary and appending internally generated subcommands.

dingtalk-minutes