reroute-plan
Pass
Audited by Gen Agent Trust Hub on Aug 16, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill instructions do not contain attempts to bypass safety filters or override agent behavior. The language is focused on establishing a governance process for re-planning, enforcing mandatory user approval ('STOP') before proceeding to different phases.- [INDIRECT_PROMPT_INJECTION]: The skill reads from a local scratchpad file to maintain state and prevent infinite loops, representing a minimal attack surface.
- Ingestion points: Reads
.skills/<CODE>/corrections.mdin Phase 2 to check for previously acted-on evidence. - Boundary markers: Not explicitly defined for the local ledger file.
- Capability inventory: The skill is limited to delegating tasks to other internal sub-skills (e.g.,
plan-tasks,design-solution) and lacks autonomous file-write or network capabilities. - Sanitization: Not applicable as the data is used for internal logic checks within the local workspace.- [COMMAND_EXECUTION]: No shell commands, subprocess calls, or dynamic context injection patterns (
!command) are present in the analyzed files.- [DATA_EXFILTRATION]: No network operations (e.g., curl, wget) or attempts to access sensitive system files (e.g., SSH keys, credentials) were found. The skill only references project-specific documentation and metadata files.- [OBFUSCATION]: The instructions and configuration are written in plain text with no evidence of encoding (Base64, Hex), zero-width characters, or homoglyph attacks.
Audit Metadata