duck-adventure
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a local Python engine (
hooks/adventure.py) to process game mechanics such as combat and movement. This execution is limited to a provided script and does not involve downloading external binaries. - [INDIRECT_PROMPT_INJECTION]: The skill handles user commands and narrates game events based on external data files (
assets/*.md). It implements multiple layers of protection, including an explicit opt-in gate, a structured command parser, and instructions for the agent to maintain the "dungeon voice" without affecting other workflows. - Ingestion points: User-supplied game commands (e.g., "attack", "move") and narrative data from local markdown assets.
- Boundary markers: Requires explicit activation signals ("duck adventure") and one-time scope approval for file access.
- Capability inventory: Execution of a local Python script and writing state to a specific hidden directory (
.duck-adventure/). - Sanitization: Instructions require the agent to scan state content for secrets or PII and redact them before writing to the state file.
- [SAFE]: The skill is strictly decoupled from productivity tasks and includes a clear exit mechanism ("done"). It operates within a strictly defined local directory and follows best practices for state management and secret handling.
Audit Metadata