# Junction

> <!-- junction:mcp start -->

- Skill: `rynaro/junction` (Agent Skill)
- Install (CLI): `npx skillmds@latest add rynaro/junction`
- Raw SKILL.md: https://api.skillmd.com/api/skills/rynaro/junction/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: Rynaro (https://skillmd.com/u/rynaro)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/rynaro/junction

---

<!-- junction:mcp start -->
# Junction MCP Skill

This skill is auto-generated by `junction mcp install --with-skill`.

## What is Junction?

Junction is the ECL (Eidolons Communication Layer) production harness. It
orchestrates multi-Eidolon hand-offs via envelope-based dispatch.

## Available MCP Tools

### harness.plan_from_prompt

Reserved for future use. In the two-phase Junction model the host LLM (Claude
Code) is the planner; Junction executes the plan. For now, construct a
plan.json manually (see the Junction spec §7.5) and call harness.run.

**Input:** `prompt` (string) — natural-language task description.

### harness.run

Execute a Junction plan. Reads plan.json at plan_path, dispatches each step
via the appropriate executor, and returns the thread_id and trace_root.

**Input:** `plan_path` (string) — absolute or relative path to plan.json.

### harness.verify

Verify an ECL envelope against L1 (schema), L2 (integrity), L3 (edge), and
L4 (performative) checks.

**Input:** `envelope_path` (string) — path to the ECL envelope file.

**Output:** `ok` (bool) + `errors` (list).

### harness.inject

STUB (v0.1). Human-in-the-loop envelope injection (§5.7/§6.5). Not
implemented yet — returns `ok: false` with a descriptive error.

**Input:** `thread_id` (string), `envelope` (object).

## Usage

The server is launched automatically by Claude Code as a stdio subprocess
via the `.mcp.json` entry written by `junction mcp install`.

To remove this wiring: `junction mcp uninstall`.
<!-- junction:mcp end -->