The Agent Skills Directory

[SAFE]: No security issues or malicious patterns were identified in the skill instructions. The skill's architecture is focused on text annotation and explicitly mandates human-in-the-loop verification before any edits are finalized.
[DATA_EXPOSURE]: The skill reads local project context such as glossaries, project notes, and specifications (e.g., AGENTS.md, CLAUDE.md) to ground its vocabulary. This is standard functional behavior for specialized agents and does not involve unauthorized access to credentials or external data transmission.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, potentially untrusted transcription data. However, the risk is negligible as the skill lacks dangerous capabilities (network access, code execution) and requires the user to manually pass through and confirm all findings.
Ingestion points: Processes auto-transcribed meeting notes or audio transcripts provided by users.
Boundary markers: Not explicitly defined for input; however, the skill uses specific markdown callouts for output.
Capability inventory: Limited to text processing and reading local project documentation. No network operations, shell execution, or dynamic evaluation patterns found.
Sanitization: No explicit sanitization of input, but the workflow requires user validation of every proposed change, providing a human firewall.

audit-transcription