edit-conversion-safe-blogs
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses a local Python script,
scripts/conversion_guard.py, to snapshot and verify the integrity of Markdown files. The script is invoked with standard arguments to track changes in links, headings, and specific 'protected' blocks of text. The execution is limited to local file operations within the user's project context. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided blog drafts. These drafts could potentially contain instructions aimed at overriding the agent's guidelines (e.g., hidden prompts to delete specific links or change the tone in unauthorized ways).
- Ingestion points: The agent reads and processes draft content from
draft.mdor text provided in the prompt. - Boundary markers: The workflow uses
<!-- conversion-safe -->HTML comments to mark sections the agent should not modify, which helps delimit content but does not act as a security sandbox. - Capability inventory: The agent has the ability to read files, write edited drafts, and execute a local helper script for verification.
- Sanitization: Integrity is maintained through a verification script that compares the edited output against a structural 'contract' generated from the original draft.
Audit Metadata