addicted2crypto

DCGP — AI skill for Claude Code

AI community

Context governance kernel for LLM agents.

How to install DCGP

This entry records only its repository, not the path inside it, so there is no exact command to give. Open addicted2crypto/DCGP and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What DCGP does

Context governance kernel for LLM agents. Predicts entropy, blocks hallucinations. Pairs with opencode-dcp. llm, llm-agents, opencode, opencode-plugin, context-governance, anti-hallucination, claude-code, cursor, typescript.

Alternatives in AI

Loushang — AI-native agent harness for coding workflows by python: multi-model LLM orchestration, stateful sessions, tool 1.3k ★
OpenContext — A personal context store for AI agents and assistants—reuse your existing coding agent CLI (Codex/Claude/OpenC 1.1k ★
Browser Search — A skill for AI agents: search the web with SearXNG, browse with Camofox, bypass protections with CloakBrowser 499 ★

README

DCGP - Dynamic Context Guidance Paths

**The Semantic Operating System for LLM Agents** · v1.0.0-rc · MIT

[![Spec](https://img.shields.io/badge/spec-DCGP--1.0-0f6e56)](./DCGP-SPEC.md) [![Conformance](https://img.shields.io/badge/conformance-FULL%2BEXTENDED-0f6e56)](./COMPLIANCE) [![License](https://img.shields.io/badge/license-MIT-blue)](./LICENSE) [![Zero deps](https://img.shields.io/badge/@dcgp/core-0%20runtime%20deps-success)](./packages/core/package.json)

DCGP is a closed-loop governance layer that eliminates context decay and hallucinations in long-running LLM sessions. It monitors context health per turn and orchestrates interventions, predicting degradation before the model begins to hallucinate.

DCGP also ships **`@dcgp/vibe-audit`** for static analysis of AI-coded ("vibe-coded") repositories - 8 rules that catch the artifacts AI-assisted code commonly leaves behind (stub markers, hardcoded credentials, type-safety bypasses, command injection, test theater, predictable randomness, ReDoS risk, comment-density imbalance). Runtime governance via the 7-step loop; static governance via `dcgp audit`. Same severity tier system, same `Finding` shape, same JSON / SARIF / Markdown formatters.

**Live state of this repo is defined by `./scripts/verify-dcgp.sh`.** It is the single source of truth for what is implemented. The prose below describes intent; the script enforces reality.

**Relationship to DCP ([@tarquinen/opencode-dcp](https://github.com/Opencode-DCP/opencode-dynamic-context-pruning)):** DCP manages context *quantity* (token budget, rule-based pruning of stale outputs). DCGP manages context *quality* (domain grounding, entropy prediction, hallucination blocking). Run both, they are orthogonal and additive.

DCGP does not reimplement DCP's pruning. DCGP emits a `RetentionDirective` on every entropy-level transition (see [Pruning Nexus](#the-pruning-nexus)); `@dcgp/opencode` translates each directive into a DCP-shaped config patch (`turnProtection`,