Outsourcerer

Přesouvá rutinní práci na levnější engine, zatímco Claude orchestrates

od alexgreensh · alexgreensh/outsourcerer

Tested · Didn't pass

Outsourcerer — Přesouvá rutinní práci na levnější engine, zatímco Claude orchestrates

Co umí

Orchestration copilot, který delegates mechanical work na cheaper backends (OpenRouter, Codex, Gemini, Devin, local models), zatímco Claude keeps the judgment calls, s cost ledger, cloud-consent gate a background job watchdog. Spouští se na outsource, offload, delegate, conserve tokens nebo asking for second opinion.

Testovací report

Tried to delegate real job — write pytest tests pro fizzbuzz module — a could not get single lane k run: devin, gemini, droid a cursor-agent nejsou installed, there is no OpenRouter key, local codex turned out to be shell alias spíše než PATH binary, a claude-native fallback died na 401, protože headless auth je separate z interactive login. Takže central claim, že cheaper engine does the grunt work, je unmeasured zde spíše než confirmed nebo denied. What did run je genuinely careful: cloud disclosure fired before anything left the machine, a source deliberately extracts one named key z ~/.env s commentem explaining, že sourcing the file would leak every other secret na third-party model. Two real gaps though — s no OpenRouter key advertised live-benchmark model picking collapses do nine-way tie at score 55, a advise then recommended the codex lane, které its own doctor had already reported jako not ready.

Testováno: 2026-07-17 · Claude Code 2.x (agent harness)

Instalace

git clone https://github.com/alexgreensh/outsourcerer.git
mkdir -p ~/.claude/skills
cd outsourcerer && cp -r plugins/outsourcerer/skills/outsourcerer ~/.claude/skills/outsourcerer

Příkazy a ukázkové prompty

  • /outsourcererPřesouvá rutinní práci na levnější engine, zatímco Claude orchestrates

Skilly se spouštějí běžnými požadavky — žádné příkazy k zapamatování. Po instalaci ho aktivují prompty jako tyto (anglicky):

  • Delegate writing these pytest tests to a cheaper model while you review
  • Can you offload this grunt refactor to Codex and just check the diff after
  • I want a cost ledger before you hand this task off to another model