plan-dead-code
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONREMOTE_CODE_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from a codebase to perform audits, creating a surface for indirect prompt injection.
- Ingestion points: The skill reads files throughout the repository (
src/**,package.json, config files, etc.) using tools likeknip,ripgrep, andjscpd(detected inSKILL.mdandreferences/residue-greps.md). - Boundary markers: There are no explicit boundary markers or instructions to ignore instructions embedded in the processed data.
- Capability inventory: The agent has the capability to execute shell commands (
npx knip,rg,jscpd,supabase-cli) and perform file system analysis (detected inSKILL.mdandreferences/residue-greps.md). - Sanitization: No explicit sanitization or filtering of the processed content is specified before it is analyzed by the LLM.
- [COMMAND_EXECUTION]: The skill heavily relies on the execution of shell commands to gather metrics and perform audits.
- Evidence: Instructions include running
npx knip,npx jscpd,ripgrep(rg),find, andsupabase-clicommands (detected inSKILL.mdandreferences/residue-greps.md). - Context: These are standard developer tools, but their use on an untrusted codebase via an agent represents a capability that could be abused if the agent is misled by malicious content within the repo.
- [REMOTE_CODE_EXECUTION]: The skill includes instructions to build the project, which can trigger arbitrary code execution.
- Evidence:
Phase 3andreferences/residue-greps.mdinstruct the agent to runnpm run buildto check for bundle leaks. - Risk: Running build scripts in an untrusted repository can execute arbitrary code defined in
package.json(e.g.,prebuildorpostbuildhooks). - [DATA_EXPOSURE]: The skill actively searches for sensitive information, though it includes safety instructions.
- Evidence: The skill searches for
service_roleandSUPABASE_SERVICE_ROLEin thedistfolder and scans.envfiles (detected inreferences/residue-greps.md). - Mitigation: The skill explicitly warns: "Report names only — never print values." and directs the agent to a specialized secrets audit skill if findings occur, demonstrating an awareness of security best practices.
Audit Metadata