goal-crafting
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data such as bug reports, support tickets, and arbitrary project context to generate executable session goals.
- Ingestion points: The 'Input' section of SKILL.md specifies that the agent should process rough tasks, bug reports, feature ideas, and partial context from project tickets or error logs.
- Boundary markers: The instructions do not define explicit delimiters or 'ignore embedded instructions' markers to wrap around the untrusted data during the processing phase.
- Capability inventory: While the skill primarily generates text output (markdown), the resulting '/goal' command is intended for the Claude Code CLI, which possesses high-agency capabilities including file system modification, command execution, and network operations.
- Sanitization: The skill includes a 'Guardrails' section that explicitly forbids the agent from pre-authorizing irreversible actions (e.g., deleting data, pushing code) and requires a 'Validate quality' step to ensure goals are terminable and focused, which acts as a logical filter against some malicious injection patterns.
Audit Metadata