[Resource]: Claude's AI Buddies

[cukas]

Awesome Claude Code Submission — Copy-Paste Fields

Submit at: https://github.com/hesreallyhim/awesome-claude-code/issues/new/choose

---

Display Name

Claude's AI Buddies

Category

Tooling

Sub-Category

Tooling: Orchestrators

Primary Link

https://github.com/cukas/claudes-ai-buddies

Author Name

cukas

Author Link

https://github.com/cukas

License

MIT

Description

A Claude Code plugin that lets you route a task to multiple AI engines — Codex, Gemini, and Claude — from inside Claude Code. Each engine gives a confidence score and proposes a technical approach, so you can compare before choosing who takes the task. Claude doesn't just relay the bids — it calibrates them, adjusting for overconfidence or missed edge cases, and recommends the best fit. Lightweight subprocess architecture that leverages your existing local CLI environments without additional API keys or MCP configuration. Also includes direct delegation (/codex, /gemini) and second-opinion code review commands (/codex-review, /gemini-review).

Validate Claims

1. Install one or both peer engines: npm install -g @openai/codex and/or npm install -g @google/gemini-cli.
2. Authenticate: codex auth login and/or gemini auth login.
3. Install the plugin: claude plugin marketplace add cukas/claudes-ai-buddies && claude plugin install claudes-ai-buddies@cukas.
4. Start a new Claude Code session. A startup banner confirms detected engines: [AI Buddies] Ready — Codex <version> Gemini <version>.
5. Run /brainstorm "write a function that validates email addresses". Each engine returns a confidence score with rationale. Claude calibrates and presents a comparison table.
6. Run /codex-review or /gemini-review with uncommitted changes to see a second-opinion code review.
7. Run the test suite without any engines installed: bash tests/run-tests.sh — 41/41 tests pass using mock CLIs.

Specific Task(s)

Open any project with uncommitted changes. Run /brainstorm "write unit tests for the changed files" to see confidence bids from each engine. Pick one to execute. Then run /gemini-review or /codex-review to get a code review of those same changes from a different engine.

Specific Prompt(s)

/brainstorm "write unit tests for the changed files"

Additional Comments

Pure subprocess architecture — no MCP servers, no API keys configured in or transiting through the plugin. The plugin itself makes zero network calls; it invokes the Codex and Gemini CLIs which handle their own auth and communication. No telemetry, no system file modifications, no bypass-permissions required. Works with any combination of engines (just Codex, just Gemini, or both). Optional config at ~/.claudes-ai-buddies/config.json for model overrides and timeouts. 41/41 tests, ~1,500 lines, all Bash + Markdown.

[hesreallyhim]

This submission was not made through the required web form. As noted in CONTRIBUTING.md, recommendations must be submitted using the web form.

A cooldown of 7 days has been applied. Please use the web form for your next submission.