ix-opencode-plugin

An OpenCode plugin that brings Ix Memory's graph-first reasoning into OpenCode as a native cognitive layer.

OpenCode + Ix = reasoning engine + persistent code knowledge graph. Skills are cognitive abstractions — not CLI wrappers — that start with cheap graph signals before reading source, and stop early when the question is answered.

Requirements

• OpenCode installed
• Ix Memory CLI (ix) installed and connected to a workspace
• Bun installed — required (all tools use Bun's $ shell API; Node.js is not supported)

opencode --version
command -v ix && ix status
bun --version

Installation

# From your project root
mkdir -p .opencode/plugins .opencode/tools .opencode/runtime .opencode/commands .opencode/agents

cp -r /path/to/ix-opencode-plugin/plugins/.  .opencode/plugins/
cp -r /path/to/ix-opencode-plugin/tools/.    .opencode/tools/
cp -r /path/to/ix-opencode-plugin/runtime/.  .opencode/runtime/
cp -r /path/to/ix-opencode-plugin/commands/. .opencode/commands/
cp -r /path/to/ix-opencode-plugin/agents/.   .opencode/agents/
cp    /path/to/ix-opencode-plugin/AGENTS.md  .opencode/AGENTS.md

Add to opencode.json:

{
  "plugin": [".opencode/plugins/ix-plugin.ts"],
  "instructions": [".opencode/AGENTS.md"]
}

The runtime/ directory is required — tools import the Ix Core Runtime client from ../runtime/client.ts.

See QUICKSTART.md for full setup instructions.

Skills

Eight slash commands with phased reasoning — graph first, source reads only when needed.

Command	What it does
/ix-understand [target]	Build a mental model of a system or the whole repo
/ix-investigate <symbol>	Deep dive: role, connections, execution path
/ix-impact <target>	Change risk: blast radius, affected systems, test targets
/ix-plan <targets...>	Risk-ordered implementation plan
/ix-debug <symptom>	Root cause analysis from symptom to candidates
/ix-architecture [scope]	Design health: coupling, smells, hotspots
/ix-docs <target> [--full]	Generate narrative-first system documentation
/ix-help [question]	Route to the right skill or tool

Agents

Agent	Purpose
ix-explorer	General-purpose graph exploration, open-ended questions
ix-system-explorer	Full architectural model of a codebase or region
ix-bug-investigator	Autonomous investigation from symptom to root cause candidates
ix-safe-refactor-planner	Blast radius + safe change sequencing for refactors
ix-architecture-auditor	Full structural health report with ranked improvements

Tools

17 TypeScript tools OpenCode can call during any task.

Tool	Purpose
ix-query	Graph entity lookup by name (locate + explain)
ix-neighbors	Callers, callees, imports, depends traversal
ix-impact	Blast radius and risk verdict
ix-map	Architectural overview with subsystem table
ix-ingest	Graph status and refresh trigger
ix-history	Revision, decisions, bugs (Ix Pro)
ix-docs-tool	Condensed context summary for injection
ix-locate	Text/pattern search across the codebase
ix-explain	Full symbol explanation with role and importance
ix-rank	Top symbols by dependents, callers, or members
ix-stats	Graph-wide statistics and health
ix-subsystems	Subsystem listing with hierarchy and signals
ix-inventory	Enumerate files or symbols in a path scope
ix-trace	Full execution path trace (upstream + downstream)
ix-decide	Pre-edit policy verdict (ALLOW / REVIEW / BLOCK)
ix-health	CLI and graph availability check
ix-smells	Architecture smell detection

All tools try the Ix Core Runtime HTTP API first and fall back to the ix CLI when the runtime is unavailable.

Hooks

Trigger	Hook	Effect
Any session	AGENTS.md injection	Always-on graph-first guidance
Before file edit	tool.execute.before → ix-decide	Policy verdict; surfaces risk for REVIEW/BLOCK
Before read	tool.execute.before → ix-read	Hints toward graph narrowing first
Before grep/bash	tool.execute.before → ix-intercept	Suggests ix text for search
After file edit	tool.execute.after → ix-ingest	Async graph refresh (fire-and-forget)
After ix-* tools	tool.execute.after → ix-errors	Detects stale graph signals
Plugin init	onInit	Triggers ix map if graph is empty

All hooks return { action: "allow" } — they inject context but never block actions.

Further reading

• QUICKSTART.md — step-by-step first use guide
• ARCHITECTURE.md — design decisions and component overview
• TOOL_CONTRACT.md — tool API reference and output format
• ROADMAP.md — implementation progress and upcoming work