DEDUPED Hot AI repos, turned into use cases Get the Monday 5

How to optimize an agent against your own task set with Gear

Package your own tasks in Harbor format and run Gear's Refine Skill to improve the agent on them.

Short answer: Build your task set in Harbor format by following the dataset guide, with a clear success criterion for each task; then run npm install --global rsi-gear@latest agent-hitch@latest @deepseek-ai/dsh@latest to install Gear and its runners; run hitch eval setup harbor to set up the Harbor evaluation runner. All steps.

Who it is forA developer who has a repeatable set of agent tasks with clear pass or fail criteria and wants the agent to get better at that specific work.

The problemNo public benchmark matches your workflow, so the agent's failures on your real tasks go unmeasured and unfixed.

The outcomeA harness tuned to your task set, with each version and its results recorded so you can compare changes and reuse the best harness.

What you need first

  • npm installed, to run the global install of rsi-gear, agent-hitch, and DSH
  • An agent environment such as Codex, Claude Code, or DSH that can use the bundled Refine Skill
  • A benchmark of representative tasks with clear success criteria, either an existing one or your own in Harbor format
  • Access to the models you choose for the Meta agent and the task agent

Steps

  1. Build your task set in Harbor format by following the dataset guide, with a clear success criterion for each task.
  2. Run npm install --global rsi-gear@latest agent-hitch@latest @deepseek-ai/dsh@latest to install Gear and its runners.
  3. Run hitch eval setup harbor to set up the Harbor evaluation runner.
  4. Install the Refine Skill in your agent environment and point its configuration at your task paths, target harness, and model settings.
  5. Invoke /refine where supported and run one optimization round, then review the recorded harness versions and their results.

Prompt for your agent

Use the Gear repo at rsi-gear/gear to optimize the harness of <target agent> on my task set at <task path in Harbor format>. Report the pass rate for the starting harness and for the best retained harness. Verify that the task paths resolve and that the Refine Skill connects to Gear before the first run.

Caveat. Results are only as good as the tasks, so the success criteria in your Harbor tasks must be clear enough for Gear to score them.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for rsi-gear/gear and "Optimize an agent against your own task set"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer