phase-task-verification
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands such as
git add,git rev-parse, and a CLI tool namedgs. These commands are used to manage the repository state and verify branch synchronization. - [DYNAMIC_EXECUTION]: The skill generates shell command strings by interpolating several variables (
RUN_ID,TASK_ID,TASK_NAME,COMMIT_MESSAGE). This script generation follows a predefined template and is used for routine automation tasks. - [INDIRECT_PROMPT_INJECTION]: The skill accepts parameters that may originate from external task descriptions, creating an attack surface for indirect prompt injection or command injection.
- Ingestion points: Parameters are passed into the skill's logic from the task environment (SKILL.md).
- Boundary markers: Parameters like
COMMIT_MESSAGEare wrapped in double quotes within the shell command (-m "{COMMIT_MESSAGE}"), but no internal escaping or sanitization logic is provided within the skill itself. - Capability inventory: The skill utilizes shell execution, repository modification via git, and environment state verification via
git rev-parse. - Sanitization: The instructions explicitly state "Do NOT sanitize
- escalate to orchestrator", which indicates that the skill relies entirely on the upstream orchestrator for input validation. This design choice avoids redundant sanitization but increases the impact if the orchestrator provides malicious or malformed data.
Audit Metadata