build-and-debug-duckstation
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill provides a comprehensive set of instructions and specialized scripts for building and debugging the DuckStation emulator. All referenced external resources point to the official repositories of the well-known DuckStation project. The instructions explicitly prohibit the acquisition or redistribution of proprietary firmware or copyrighted game data.\n- [COMMAND_EXECUTION]: The skill includes standard build instructions and test scripts. Evidence includes:\n
- Build Commands: The
references/duckstation-ps1-source-builds.mdfile provides standardgit,cmake, andninjacommands for building the project from source.\n - Test Script:
scripts/test_build_command.pyusessubprocess.runto execute a local Python helper script. This is used strictly for unit testing the command-line builder logic and does not incorporate untrusted external input into the shell environment.\n- [INDIRECT_PROMPT_INJECTION]: The skill involves processing output from build tools and emulators, creating a potential surface for indirect prompt injection. However, this is inherent to the skill's purpose and is well-mitigated. Evidence chain:\n - Ingestion points: Build logs, guest registers, renderer captures, and GDB server output are analyzed by the agent (found in
SKILL.mdandreferences/duckstation-ps1-verification-and-claim-evidence.md).\n - Boundary markers: The
Operating contractinSKILL.mdrequires explicit validation at the boundary of claims and mandates that the agent inspect tool output before relying on it.\n - Capability inventory: The skill uses shell commands for building and a Python script for command construction.\n
- Sanitization: The skill mandates the use of isolated task-owned directories and manual inspection of technical logs to prevent cross-contamination or unauthorized actions.
Audit Metadata