orca-orchestration

SUSPICIOUS: the skill is largely coherent with its stated orchestration purpose and uses plausible same-project tooling, but it deliberately enables unattended agent execution with disabled approval controls and autonomous git actions. This is not confirmed malware or credential theft, yet it is a high-impact automation skill whose real footprint exceeds a low-risk helper.

This code simulates an Orca orchestration CLI and stores state and argument logs in environment-selected files. Its worker-start path can execute the FAKE_ORCA_ON_START environment value as a shell command, making command execution possible when that variable is attacker-controlled and the path is invoked. No network exfiltration, credential harvesting, or destructive behavior is otherwise evident. Severity is conditional and potentially high in an environment where untrusted parties can set that variable; otherwise the fragment appears to be a test harness.

This code manages configured workflow queue items and runs optional configured shell hooks. It shows no direct evidence of malware, data exfiltration, or obfuscation. Untrusted queue YAML can execute arbitrary hook commands, and unchecked item IDs can cause path traversal in state and run-directory operations; treat the queue configuration as trusted and validate IDs. Severity is moderate when queue configuration may be attacker-controlled.

As shown, the fragment is syntactically invalid because `EvalError` has no class body. If corrected, it evaluates conditions and can execute resolved condition text through a supplied shell when no expected value is set, so untrusted condition or scope data could lead to command execution; no specific malicious payload, network destination, or credential theft is present. Severity is conditional on whether these inputs are trusted.