octocode-roast
Pass
Audited by Gen Agent Trust Hub on Aug 9, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The instructions suggest the installation of
octocode-mcpusing the commandnpx -y octocode-mcp. This process downloads and runs a package from the NPM registry, which constitutes remote code execution from an external source. - [EXTERNAL_DOWNLOADS]: The skill relies on the external
octocode-mcppackage for its core functionality, recommending its download and installation if it is not already present in the environment. - [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes content from local files that may contain untrusted data. Ingestion points: Code files are accessed via
localGetFileContentandlocalSearchCodetools in the defined project scope. Boundary markers: While the skill uses specific output templates, it does not implement strong delimiters or instructions to ignore embedded prompts within the ingested code. Capability inventory: The agent can perform local file searches, read file contents, and modify code during the redemption/resurrection phase. Sanitization: The skill explicitly requires redacting sensitive values like API keys from its output, which mitigates direct exfiltration risks but does not prevent logic-based injection attacks. - [COMMAND_EXECUTION]: The skill uses local tools and version control commands (e.g.,
git diff) to establish the scope of the code review and identify target files.
Audit Metadata