The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The launcher scripts scripts/impeccable and scripts/impeccable.cmd are designed to download platform-specific binary engines from an external GitHub repository (github.com/pbakaus/impeccable) during the first run.
[REMOTE_CODE_EXECUTION]: After downloading the binary engine, the skill assigns executable permissions and runs the binary on the host system. While the skill includes checksum verification logic, the source is an external third-party that is not recognized as a trusted organization or well-known service in the security guidelines.
[PERSISTENCE]: The skill attempts to establish automated execution hooks by modifying configuration files for AI agent environments (such as .claude/settings.local.json, .cursor/hooks.json, and .codex/hooks.json). These hooks ensure that the skill's detector logic runs automatically on every subsequent file edit within the project.
[COMMAND_EXECUTION]: The skill invokes system-level utilities for mobile device and simulator manipulation, such as adb exec-out for Android and xcrun simctl for iOS. It also interacts with local DevTools ports (9222) to inspect desktop applications.
[PROMPT_INJECTION]: Instruction files in reference/degraded/ contain meta-instructions directing the agent to disclose internal role-substitution information 'outside the output contract' and to 'step fully out of the work you just finished.' This pattern attempts to define communication constraints and role-play behaviors that could be used to minimize user oversight of agent state transitions.

impeccable