interactive-markdown-review
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to read and process external markdown documents where content within
<single brackets>is interpreted as author instructions. This creates a surface where a malicious document (e.g., in a compromised repository) could inject instructions to influence the agent's behavior. - Ingestion points: Reads markdown documents (
<document.md>) and JSON notes files (.notes.json). - Boundary markers: Uses specific bracket patterns (
<...>,<<...>>) to distinguish between author and editor input. - Capability inventory: The agent can write to the local file system, execute Node.js scripts, and run shell commands like
gitandlavish-axi. - Sanitization: The
render.mjsscript performs HTML escaping for the rendered review page. - [COMMAND_EXECUTION]: The skill relies on executing Node.js scripts (
serve.mjs,render.mjs) and external CLI tools (git,lavish-axi). While these are standard developer tools, they represent an execution capability that could be abused if the agent's logic is subverted. - [DYNAMIC_EXECUTION]: The
SKILL.mdfile contains instructions for the agent to evolve the skill by folding lessons into its ownSKILL.mdfile. This self-modification of the agent's instruction set is a form of dynamic code modification, although the protocol explicitly requires the author's agreement before making such changes. - [EXTERNAL_DOWNLOADS]: The documentation (
README.mdandvisual-mode.md) directs the agent to check for and potentially installNode.jsand thelavish-axipackage vianpm. The instructions specify that the agent should only proceed with user consent and handle permissions (likesudo) carefully.
Audit Metadata