The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The assets/examples/verify.sh script downloads DuckStation release binaries directly from the official project repository on GitHub. The download process includes a mandatory SHA-256 integrity check using a hardcoded hash to ensure the downloaded asset has not been tampered with.
[COMMAND_EXECUTION]: The skill uses shell commands and debugging utilities like lldb to launch and inspect the DuckStation process. This functionality is restricted to the isolated portable copy created by the skill, protecting the user's primary application data and settings.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data in the form of .ini configuration files and .cht cheat files. To mitigate risks associated with malformed or malicious inputs, the scripts/check_formats.py tool validates all keys and values against a strict whitelist of known-good settings for the specific release version.
Ingestion points: scripts/check_formats.py (reads user-supplied settings.ini, .cht files, and texture names).
Boundary markers: Uses portable.txt to isolate the user directory and temporary scratch folders for all diagnostic runs.
Capability inventory: Uses subprocess.run (in test scripts) and shell execution to launch the emulator binary.
Sanitization: Implements strict key-name validation against a bundled reference list and regex-based format checking for cheat codes.
[DYNAMIC_EXECUTION]: The skill's automated test suite (scripts/test_build_command.py and scripts/test_check_formats.py) uses subprocess.run to execute internal Python utility scripts. This is used exclusively for testing the skill's logic in a controlled environment.

debug-duckstation