Anti-Dark-Code: Understand and Verify Code (Preview)

When you inherit a repository or need to review AI-generated code, the first problem is knowing what is there and which claims you can trust.

What it does

Anti-dark-code guides your assistant through the work relevant to your request:

• Map important modules, entry points, data flows and trust boundaries.
• Investigate defects and risks using source evidence and reproducible checks.
• Explain code and improve documentation while preserving behavior and important identifiers.
• Choose proportionate verification, inspect test strength and report actual results.
• Propose focused repairs, and carry out changes when you authorize their scope and effects.

Reports distinguish directly verified facts from inferences and unknowns. Planned checks, executed checks and passing results remain separate. You can request plain-language explanations without removing evidence or approval requirements.

Works independently

Scaffold Kit is optional. Anti-dark-code works with your existing repository, documentation and task scope; you do not need a new planning document set or a managed local installation for a one-off review. If you use both skills, it can supply observed maps and verification evidence for Scaffold Kit's plan.

Optional bundle: Scaffold Kit + Anti-Dark-Code: Plan and Verify is $39 one-time, saving $9 against $48 separately. Both skills remain available individually.

Requirements

• Your own compatible AI agent with access to the supporting skill files and the project material you authorize it to inspect.
• Python 3.12+ for bundled helpers; Git for Git-based provenance features. Instruction-led inspection can proceed without the helpers, with missing machine checks reported.
• The project's own tools, dependencies, credentials or services for any checks that require them. A configured command is not permission to execute it.
• An AI account and any model/API usage are separate from this purchase.

Limitations and privacy

• This is a preview: version 2026.10.15-unified.21. It has not been promoted to a stable upstream release.
• An audit covers the named scope. It is not a security certification, a guarantee that every defect was found, or proof that unexecuted behavior works.
• Source-bound records help preserve traceability; they do not replace human authorization or establish that old evidence is still current.
• Optional features can modify managed local files or run reviewed commands. Optional CI retrieval can contact GitHub using your supplied credentials.
• Usage observation is off until you explicitly enable it. When enabled, it can read selected local host logs and retain usage counters and technical metadata; it does not retain transcript bodies or upload by default.
• Host and runtime support vary. Advanced managed migration also requires the original release's operator documentation.
• In Claude, creator-free Agensi access retrieved the main instructions and ten supporting text files, covering investigation, documentation and verification references, the system-map template and the main Python helper source.
• These checks do not cover buyer checkout/unlock or delivery, running Agensi-delivered helpers, every remaining file or conditional workflow, or complete byte-for-byte matching to the submitted ZIP. Agensi excludes the two bundled branding PNGs from agent text delivery.
• The example screenshot came from the exact creator-supplied ZIP in a separate Claude conversation; it does not demonstrate buyer delivery.

Source and license

The source is publicly available on GitHub and includes FSL-1.1-MIT license terms. The listing offers the packaged workflow through Agensi. https://github.com/LynxTWO/anti-dark-code-skill