outsource — use convertible as a different mind

SkillProductivity

Hand a scoped repo task to convertible — a *different* engine/model than you (e.g. a local vLLM Qwen) — and fold its answer back. The point isn't a stronger model; it's a different mind, and diversity helps: `outsource review` gets an independent second opinion on a diff, `outsource explore` gets a fresh read of an area, `outsource write` delegates a small implementation. Use when the user says "outsource this", "get a second opinion", "have convertible review/explore/ write", "ask the other model", or when you want a diverse perspective rather than just doing it yourself. Read-only verbs (explore/review) run isolated in a throwaway git worktree and cannot touch the working tree.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the outsource — use convertible as a different mind skill

What this skill tells your AI

The instructions your AI receives, as published by agentculture/reachy-mini-mcp in .claude/skills/outsource/SKILL.md and read by ahel’s review.

outsource drives the convertible CLI so a Claude agent can hand a scoped task to a different engine (default: a local vLLM Qwen3.6-27B on :8001). Convertible's model is not assumed to be stronger than you — its value is diversity. A second, independent mind catches things the author's mind glides past, which is why review is the headline verb.

This skill is the operator: a portable wrapper that resolves the CLI and turns each verb into a convertible drive, then prints the drive's result summary.

How to run

The entry point is scripts/outsource.sh. Invoke it from the repo you want convertible to work on:

bash .claude/skills/outsource/scripts/outsource.sh <verb> "<text>" [options]

It resolves the CLI portably — an installed convertible on PATH (the normal case), falling back to uv run convertible when inside the convertible checkout, else an install hint.

Verbs

VerbWhat it doesSide effects
explore "<question or area>"Read-only investigation of the repo; the model reads and reports findings.None — runs in a throwaway worktree at HEAD.
review "<what to focus on>" [--base main]A diverse second opinion on the committed diff (<base>...HEAD).None — throwaway worktree; reviews committed changes only.
write "<task>" [--pr]Implement a change. Commits to a drive branch by default; --pr pushes + opens a PR.In-place: a convertible/<id> drive branch (or a PR).

Options

OptionMeaning
--repo PATHTarget repo (default: .).
--base BRANCHBase for the review diff (default: main).
--engine NAMEEngine wheel (default: $CONVERTIBLE_ENGINE or vllm-openai).
--model NAMEModel (default: $CONVERTIBLE_MODEL or mmangkad/Qwen3.6-27B-NVFP4).
--base-url URLOpenAI base URL (default: $CONVERTIBLE_BASE_URL or http://localhost:8001/v1).
--max-steps NLoop step budget (default: 20).
--allow-dirty(write) allow running on a dirty tree.
--pr(write) push + open a PR instead of a local drive branch.

The result printed to stdout is the drive's TaskResult.summary (plus changed_files / drive branch for write), parsed from convertible drive --json. Per-step progress streams to stderr while it runs.

When to reach for which verb

  • review — the standing use. You wrote (or an agent wrote) a change and you want a candid, independent pass over the committed diff before you trust it. Treat the output as a second opinion to weigh, not a verdict.
  • explore — you want a fresh, unbiased read of an unfamiliar area ("how does X work here?") without anchoring on your own assumptions.
  • write — a small, well-scoped implementation you're happy to delegate. The result lands on a drive branch you can inspect, merge, or discard.

Hard rules (do not violate)

  • explore and review are read-only. They run in a throwaway git worktree at HEAD, so a stray write can't reach your working tree or branch; the prompts also tell the model not to modify anything. Don't route a change-making task through them — use write.
  • write refuses a dirty tree unless you pass --allow-dirty. This guards the dirty-tree hazard: convertible drive --no-pr commits uncommitted edits onto the drive branch and leaves you there. Commit or stash first.
  • Outsourced output is a second opinion, not authority. The engine may be a smaller/different model; weigh its findings, verify its claims, and own the decision yourself.

Honest limits

  • Read-only is enforced by worktree isolation + prompt constraint, not a sandbox — the loop always exposes write_file/run_command, so the model can still run arbitrary read-only commands.
  • review covers committed changes only (<base>...HEAD). To review uncommitted work, commit it first.
  • The default engine is whatever single model is running locally; a multi-model fleet (different model per verb) is separate infrastructure.

Provenance

This is a first-party convertible skill — agentculture/convertible is its origin. guildmaster re-broadcasts it to the mesh (the same inbound pattern as the devague-origin workflow skills), tracking it in docs/skill-sources.md. The cite, don't import policy holds: downstream repos copy it, they don't symlink or depend on it.

Signals

GitHub stars
32
Forks
6
Last commit
Jul 2026
Advanced
Catalog kind
skill
Gateway key
outsource
Source
github.com/agentculture/reachy-mini-mcp