Anchor Protocol
A deterministic, token-efficient project orientation protocol for human engineers and AI coding agents.
Organization: VorteXpert Labs
Author: Sabber Ahamad Emon
Why Anchor Protocol?
Modern AI coding agents frequently lose the trajectory of a software project:
- They forget architectural decisions made in previous sessions.
- They repeatedly suggest approaches that were already tested and rejected.
- They mistake local debugging rabbit holes for the overall project objective.
- They drift because context compaction erases negative constraints and boundaries.
Anchor Protocol provides a lightweight, Git-versioned layer of crystallized project orientation (.anchors/) that keeps both humans and AI agents continuously aligned.
«Context is fluid. Anchors are crystallized.»
Architecture & Lifecycle
graph TD
A[Raw AI Session Context] -->|Crystallization| B(Anchor Protocol Draft: PROPOSED)
B -->|Human/Agent Consensus| C[ACTIVE Anchor]
B -->|Flawed Premise| D[INVALIDATED Anchor]
C -->|New Strategy| E[SUPERSEDED Anchor]
C -->|Phase Exhausted| F[RETIRED Anchor]
subgraph Persistent Storage `.anchors/`
C
E
F
end
Core Philosophy
- Not AI Memory: It is not a conversational transcript, second brain, or vector dump.
- Not a Replacement for Git or Code: Code is the authority on what is implemented; Git is the authority on line diffs; Anchors are the authority on why the project is structured this way and what boundaries must not be crossed.
- Token-Efficient by Design: Uses a 3-tier progressive disclosure model (~400 tokens pre-flight overhead) optimized for modern LLM KV prefix caching.
Quickstart
1. Instant Zero-Install Usage (npx)
Anchor Protocol is published on npm as @vortexpert-labs/anchor-protocol and runs cross-platform on macOS, Linux, and Windows:
# Initialize Anchor Protocol in a new repository
npx @vortexpert-labs/anchor-protocol init --stage PROTOTYPE
# In an existing project (auto-detects tech stack and drafts baseline anchors):
npx @vortexpert-labs/anchor-protocol init --brownfield
# Validate schema integrity, acyclic DAGs, and verify zero session pollution:
npx @vortexpert-labs/anchor-protocol lint --strict
# Check or sync orientation index table:
npx @vortexpert-labs/anchor-protocol index --sync
# Fast combined pre-flight / CI check:
npx @vortexpert-labs/anchor-protocol check --strict
2. Local Repository CLI Launchers
You can also run the local repository launchers directly across any OS:
# macOS / Linux / WSL:
./bin/anchor check --strict
# Windows (Command Prompt):
bin\anchor.cmd check --strict
# Windows (PowerShell):
.\bin\anchor.ps1 check --strict
3. Universal Agent Skill Integration
Give your AI coding agent (Claude Code, Gemini CLI, Cursor, Copilot, Cline, Windsurf, Antigravity) this repository link or point it to SKILL.md:
"Add Anchor Protocol to this repository from https://github.com/vortexpert-labs/anchor-protocol"
The agent will automatically read SKILL.md, initialize .anchors/INDEX.md, and inject the non-destructive adapter into your local instruction file (AGENTS.md, CLAUDE.md, .cursor/rules/, etc.).
Platform Distribution & Guides
- Universal Agent Skill (
SKILL.md) - Full Formal Specification (APS-1)
- Research Board Report & Empirical Grounding
- Brownfield Adoption Guide
- Platform Publishing & Distribution Guide
License
Apache License 2.0. See LICENSE for details.