---
name: anchor-protocol
description: Persistent project orientation and trajectory alignment protocol for AI coding agents. Enforces pre-flight orientation, active constraints, rejected path avoidance, and progressive architectural crystallization without session history pollution.
version: 1.0.0
license: Apache-2.0
author: VorteXpert Labs (https://github.com/vortexpert-labs)
compatibility: ["claude-code", "gemini-cli", "antigravity", "cursor", "copilot", "cline", "windsurf", "universal-agents"]
metadata:
  tier: 0
  cache_optimization: prefix_preserving
---

# Anchor Protocol Universal Agent Skill

**Organization:** [VorteXpert Labs](https://github.com/vortexpert-labs)  
**Author:** [Sabber Ahamad Emon](https://github.com/sabber-ahamad-emon)

## 1. Core Purpose
The Anchor Protocol provides persistent, crystallized project orientation across multi-session AI development. It prevents goal drift, decision amnesia, and zombie re-exploration while keeping prompt overhead bounded to $\mathcal{O}(1)$ (~400 tokens) via progressive disclosure.

## 2. Epistemic Invariants
- **Code** is the authority on *what is currently implemented*.
- **Anchors (`.anchors/`)** are the authority on *project stage, goals, active constraints, and negative boundaries*.
- **Rule of Supremacy:** Never modify working code to match a stale anchor without explicit human confirmation.

## 3. Universal Bootstrap Protocol (When Asked to Install/Init)
When instructed by a user to *"Add/integrate Anchor Protocol from https://github.com/vortexpert-labs/anchor-protocol"*:

1. **Check for Existing Repository Assets:**
   - If `.anchors/INDEX.md` is absent, create `.anchors/` and copy/generate `schema.json` and `INDEX.md`.
   - In existing codebases, scan manifest files (`package.json`, `go.mod`, `Cargo.toml`, `pyproject.toml`) and establish 3–5 initial anchors (`STAGE`, `DECISION`, `CONSTRAINT`, `BOUNDARY`).
2. **Inject Platform Adapter Non-Destructively:**
   - Identify the host agent environment and inject the standard delimited adapter block into the appropriate file (`AGENTS.md`, `CLAUDE.md`, `.cursor/rules/anchor-protocol.mdc`, `.github/copilot-instructions.md`, `.windsurfrules`, or `.clinerules`):

```markdown
<!-- ANCHOR-PROTOCOL-ADAPTER: START -->
## Project Orientation & Anchor Protocol
This repository uses the **Anchor Protocol** for persistent project orientation.
- **Canonical Index:** `.anchors/INDEX.md`
- **Pre-flight Sequence:** Inspect `.anchors/INDEX.md` before executing multi-file architectural changes.
- **Invariant Rules:** Obey all `ACTIVE` constraints and never re-attempt `REJECTED_PATH` anchors.
- **Crystallization:** When establishing a consequential architectural decision or discovering a fatal flaw in an approach, draft an anchor in `.anchors/ANC-XXXX-[slug].md` with `status: PROPOSED`.
- **Anti-Pollution:** Never write raw conversation transcripts, bug stack traces, or temporary logs to `.anchors/`.
<!-- ANCHOR-PROTOCOL-ADAPTER: END -->
```

## 4. Standard Agent Operating Procedure (AOP)

### Phase 1: Pre-flight Orientation
1. Inspect `.anchors/INDEX.md` before generating plans or multi-file edits.
2. Ingest `INDEX.md` to identify:
   - Current **Project Stage** (`PROTOTYPE`, `ALPHA`, `PRODUCTION`, `MAINTENANCE`).
   - Active **Goals** and non-negotiable **Constraints**.
   - Prohibited approaches (`REJECTED_PATH`).
3. If user prompts conflict with an `ACTIVE` constraint, flag the conflict immediately and request guidance before writing code.

### Phase 2: In-flight Execution & Targeted Consultation
1. When modifying a subsystem governed by a specific anchor, read that targeted `ANC-XXXX.md` file using file-reading tools.
2. Do NOT read all anchor files by default; utilize progressive disclosure to minimize context bloat.

### Phase 3: Crystallizing Consequential Anchors
When an architectural decision, project stage transition, or failed technical approach is established:
1. Apply the **4-Question Litmus Test**:
   - Is it durable across $>3$ weeks?
   - Is the rationale invisible from code syntax alone?
   - Does it prevent future agents from repeating a failed path?
   - Can it be stated in 1–3 dense invariant propositions?
2. If YES, draft an anchor in `.anchors/ANC-XXXX-[slug].md` with:
   - `status: PROPOSED` (or `ACTIVE` if user confirmed)
   - `authority: AGENT_PROPOSED`
   - Concise claims and rationale.
3. Update `.anchors/INDEX.md` with the new anchor entry.

### Phase 4: Anti-Patterns ("What NOT to Anchor")
- Do NOT record raw session transcripts or journals ("Today we ran tests...").
- Do NOT record bug stack traces or temporary typos.
- Do NOT duplicate code syntax or schemas in markdown.
- Do NOT copy generic third-party library manuals.