project-review
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest implementation reports from ticket comments and re-run the commands contained within them to verify project deliverables. This creates a surface where an attacker can provide malicious commands via the ticket tracker that the agent will subsequently execute.
- Ingestion points: Implementation reports from ticket comments are processed as described in the 'Implementation report intake' section of references/WORKFLOW.md.
- Boundary markers: The protocol uses structured records (Charter, Evidence) but does not isolate the command verification step from the external data source.
- Capability inventory: The agent is expected to execute terminal commands to verify software and skill behavior as part of the review profiles.
- Sanitization: The instructions do not define any sanitization or validation logic for the commands retrieved from the reports.
- [COMMAND_EXECUTION]: The workflow involves executing shell commands derived from producer-supplied reports for verification purposes. This capability poses a risk if the reports originate from compromised or untrusted sources.
Audit Metadata