Sidekick — AI Coding Agents for Claude Code

Coding agents as Claude sidekicks — each agent auto-installs, configures itself, and lets Claude delegate implementation work while focusing on planning and communication.

Sidekicks

Sidekick	Skill	Agent	Status
Forge	forge	ForgeCode — #2 Terminal-Bench 2.0 (81.8%)	✅ v1.2.2

More sidekicks planned.

---

Installation

Add to ~/.claude/settings.json:

{
  "extraKnownMarketplaces": {
    "alo-exp": {
      "source": {
        "source": "github",
        "repo": "alo-exp/sidekick"
      },
      "autoUpdate": true
    }
  },
  "enabledPlugins": {
    "sidekick@alo-exp": true
  }
}

On the next Claude session, all sidekicks install automatically.

---

Forge — ForgeCode Sidekick

What it does

• Auto-installs ForgeCode on first session start
• Guides OpenRouter API key setup (Qwen3 Coder Plus — best coding model, $0.33/$1.95/MTok)
• Turns Claude into an orchestrator: Claude plans and communicates, Forge executes all file changes, tests, and commits
• Fallback ladder: automatic 3-level recovery on failure — L1 Guide (reframe), L2 Handhold (decompose), L3 Take over (Claude acts directly + DEBRIEF)
• AGENTS.md mentoring: after every task, Claude extracts learnings and writes them to a 3-tier instruction store (~/forge/AGENTS.md, ./AGENTS.md, session logs) — delegation gets smarter over time
• Skill injection: 4 bootstrap skills (testing-strategy, code-review, security, quality-gates) auto-injected into task prompts based on task type
• Token optimization: task prompts capped at 2,000 tokens with validated .forge.toml compaction defaults

How it works

You → Claude (plan + communicate) → Forge (implement + commit) → Claude (review + report)

Claude handles: architecture, explanations, research, code review Forge handles: writing files, features, tests, git commits

After installation

Claude will guide you to:

1. Sign up at openrouter.ai (Google/GitHub OAuth, ~30 seconds)
2. Add $5 credits at openrouter.ai/settings/credits
3. Create an API key and paste it into Claude

Claude configures Forge automatically and delegates all coding work from that point.

Providers & Models

Provider	Model	Notes
OpenRouter (recommended)	Qwen3 Coder Plus qwen/qwen3-coder-plus	Default. 1M context, tool-use, $0.33/$1.95/MTok
OpenRouter	Gemma 4 google/gemma-4-31b-it	Alternative. Dense, fast, separate rate limit
MiniMax Coding	MiniMax M2.7 MiniMax-M2.7	Direct API — get key at platform.minimaxi.com

---

Testing

Three-tier pyramid. All three stages are chained by tests/run_release.bash, which is the gate every release must pass.

Tier	Script	Runs without Forge	Purpose
Unit + integration	tests/run_all.bash	✅	13 suites, ~80+ assertions — hook classifiers, idx audit, plugin integrity, slash commands, v1.2 coverage gaps.
Smoke	tests/smoke/run_smoke.bash	skip	forge --version + trivial forge -p round-trip against the real binary.
Live E2E	tests/run_live_e2e.bash	skip	Full Claude→Forge delegation on a seeded-buggy testapp (tests/testapp/) — proves the 5-field prompt shape, tool-use, and verification loop work end-to-end.

Stages 2 and 3 are gated behind SIDEKICK_LIVE_FORGE=1 so they never run in CI. Before tagging a new version:

SIDEKICK_LIVE_FORGE=1 bash tests/run_release.bash

Without the env var the gate still runs stage 1 and cleanly skips 2+3 (exit 0), so it's safe to wire into CI.

---

License

MIT — Ālo Labs