unknowbug/anchorlaw
Code verification protocol for vibe coding — every claim must have a verifiable practice anchor.
A code-verification toolchain for AI-assisted (vibe) coding, maintained as a DeepSeek Harness host adaptation: 11 anchor-* protocol skills, 4 model tools (anchorlaw_scan, anchorlaw_report, anchorlaw_ai_context, anchorlaw_status), and the anchorlaw agent preset — a judge-driven four-stage pipeline (input contract → implementation spec → plan → implementation → delivery) with acceptance criteria first, a 3-round hard stop, and confirmed granted only by the human. Language-neutral protocol core (spec + python + typescript) drives the DSH tools; scanner detects P1-P6 defensive patterns with per-language annotation extraction (cpp/go/java). Components marked EXPERIMENTAL/UNVERIFIED are honestly labeled working hypotheses.
Install
pwsh dsh/scripts/install.ps1Script install per README (English, DSH host adaptation): clone the repo, then pwsh dsh/scripts/install.ps1 (host-level default: preset + user skills + global tool mount), verify with pwsh dsh/scripts/selfcheck.ps1, then open a NEW DSH session — 4 anchorlaw_* tools and 11 anchor-* skills appear. Per-project install: pwsh dsh/scripts/install.ps1 -Project /path/to/project. Tools appear only in new sessions (session composition is fixed at creation); the global mount is gated by a tool-schema check.
Compatibility
DeepSeek Harness host (maintained). PowerShell (pwsh) required for the installer. Python backend (anchorlaw-scanner, verified) + TypeScript scanner in development.
Details
- Repo: unknowbug/anchorlaw
- Category: Coding & Development
- Stars: 4
- Version: GitHub source (protocol v0.19, DSH adaptation v0.18)
- Last push: 2026-09-17
- First seen: 2026-06-17
Recent updates
v0.18/v0.19: DSH host adaptation (11 skills, 4 tools, preset, host-level global mount); verification-scope clarification; archived Reasonix format; 4 new tests; protocol v0.19.
FAQ
- How is it installed?
- One script: pwsh dsh/scripts/install.ps1, then pwsh dsh/scripts/selfcheck.ps1; open a new DSH session to see the tools.
- What does the anchorlaw preset do?
- It is a judge-driven pipeline: input contract → implementation spec → plan → parallel implementation → delivery, with acceptance criteria first and a 3-round hard stop; only the human grants confirmed.
- Is everything verified?
- No — the README honestly labels components: the Python scanner is VERIFIED on real projects; anchors/noise cards/AI-context are EXPERIMENTAL to CONJECTURE.
Alternatives
jinguanghai/deepseek-harness-forge-plugins#evidence-first · cirelir/dsh-change-review · PerryLink/dsh-auto-review