build-dark-factory

BENIGN for purpose alignment but HIGH RISK operationally. The skill's broad powers—writing repo automation, installing triggers, and enabling unattended auto-merge/deploy—are consistent with building a dark factory, and there is no clear credential theft or malicious exfiltration. However, it enables autonomous real-world actions in the user's actual repo and executes repo-local automation scripts, so the security risk is high even without evidence of malware.

No clear evidence of intentional malware within this fragment (no exec/eval, no direct network or credential theft code). The dominant concerns are security-hardening issues: (1) argv-derived paths are used directly for filesystem reads/writes/append, enabling potential path traversal and unintended file modification if attacker-controlled inputs reach this CLI; (2) sys.path manipulation broadens the import surface for gh_backend, creating a module shadowing/supply-chain sabotage risk. Additionally, parse() contains a likely logic error that may affect correctness, increasing the chance of unintended metadata handling. Review invocation context/permissions, implement strict path normalization/enforcement within ROOT, and avoid or harden sys.path modifications when importing gh_backend.

This file is a powerful CI gate/orchestrator. It does not show explicit malware actions in the fragment (no network/file theft/credential theft), but it creates a high supply-chain risk surface by executing command strings from harness.config.json via subprocess.run and by importing/running local repository Python code after modifying sys.path. If an attacker can tamper with harness.config.json or the local scripts/modules it invokes, this harness can become an execution primitive.

This fragment appears to be a CI/test harness orchestration utility. It does not show explicit malware behaviors (no exfiltration to external networks, no persistence, no obfuscated payloads, no eval/exec). The dominant security concern is high-impact, configuration-driven arbitrary process execution via subprocess.Popen/subprocess.run (for HTTP start, CLI smoke tests, and import-check), plus sys.path.insert(ROOT) which can increase the impact of unexpected or malicious code under the repository root. If configuration or repository contents are not fully trusted, this becomes a plausible CI/test-time sabotage vector. Additionally, early failure paths can include a child stdout excerpt in exceptions, which may leak secrets into logs.