automate-me
Pass
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill analyzes historical agent transcripts to generate new instructions for personalized modes.
- Ingestion points: Reads transcript files from the workspace-scoped agent-transcripts/ directory as instructed in Step 1.
- Boundary markers: The skill instructs subagents to identify signals but lacks specific technical delimiters or escaping instructions to separate untrusted transcript content from the agent's internal logic.
- Capability inventory: The skill can write files to the system via the create-skill tool and execute Git commands for repository management.
- Sanitization: Implements a cross-checking strategy requiring patterns to appear in multiple history slices before they are codified, providing a basic filter against anomalous or adversarial data in the history.
- [COMMAND_EXECUTION]: The skill utilizes shell commands to inspect the environment and manage the repository.
- Employs git log to check the modification timestamps of existing skill files in the project and home directories.
- Uses Git for branch management, commits, and opening pull requests during the landing phase.
Audit Metadata