minimax-ai/mcode-stack

harness-experiment

Run a harness change as a controlled experiment, with a small end-to-end eval, one change at a time, repeated runs, and looking at flipped tasks, request count, tokens and cache hits, not just the average score.

View source

How to use

  1. Copy the installation command and review the source package before installing it in your own agent environment.
  2. Check the source document for required tools, credentials and supported hosts.
  3. Describe your input and desired deliverable; review the output against the source material.
npx skills add https://github.com/minimax-ai/mcode-stack --skill harness-experiment

Before you start

  • Installation does not include a model subscription, API allowance or permission to access your files.
  • Repository stars and installation counts describe activity, not an independent quality or security test.
Original skill document

Rendered from the source repository. Headings, examples, code, tables, links, and referenced images are preserved.

Harness Experiment

Harness engineering moves forward by experiment. The principle: measure first, change one thing at a time, keep what survives testing.

Companion to MCode Engineering Notes 01, 02 and 05.

1. Before changing anything

  1. State the hypothesis in one sentence: what you change, what you expect

to improve, and what might get worse.

  1. Pick or build a small end-to-end eval. A handful of golden tasks is

enough to start. Include tasks for every chain the agent runs in production and every rule it must not break. The checks must not be written by the agent under test; tests an agent writes for itself share its blind spots.

  1. Pin everything else. Same model version, same settings, same tool set,

same repository state. If the model itself changes, treat that upgrade as the one change and rerun the same eval.

  1. Record a baseline. Run each task several times; agent runs are noisy and

one run can hide a failure.

2. Make exactly one change

One prompt edit, one tool moved, one subagent merged, one threshold changed. If you need several, run several experiments.

3. Measure the same things on both sides

For baseline and candidate, record:

MetricWhy
Pass rate per task (over repeated runs)Quality, with noise visible
Flipped tasks (pass to fail, fail to pass)Averages hide regressions
Model requests per taskMany wins come from fewer round trips
Input and output tokensCost and context pressure
Cache hit rateA change can save tokens but break caching
Wall-clock timeLatency the user feels

Only report numbers you actually measured, with run counts.

4. Read the result

  • Open every flipped task and read the trajectory, not just the final answer

(see trajectory-review). Find out why it flipped.

  • A change that keeps the score flat but cuts requests or tokens can be a win.

A change that raises the average while breaking a must-not-break rule is not.

  • If the evidence is mixed, keep the simpler version.
  • Write the result down: hypothesis, change, numbers, flipped tasks, decision.

Next to the code (a PR description or an experiments log) is best.

5. Re-check after model upgrades

Scaffolding built around a model's weak spots ages fastest. After each model upgrade, rerun the eval and ask which parts of the harness still earn their keep.

Output

A short experiment record: hypothesis, the single change, eval set and run count, the metric table for baseline and candidate, flipped tasks with causes, and the decision (keep / revert / iterate).

from this repository

More skills

All skills
minimax-ai
Community

engineering-code-review

Use when reviewing code, pull requests, patches, CLs, diffs, or proposed implementations for engineering quality, code health, design, functionality, tests, maintainability, specialist risk, or approval risk. TRIGGER on "review this PR", "code review", "LGTM?", "approve?", "is this code/diff/change safe to merge/deploy?", and local diff reviews. DO NOT TRIGGER for PR descriptions, PR splitting, author review-feedback responses, or non-code safety questions unless code review is requested.

1h installs
1
GitHub stars
685
Updated
Nov 10
minimax-ai
Community

evidence-first-review

Review an agent's finished work by reading the evidence before its summary, naming what each check proves and what it misses, pulling out the decisions it made silently, and trying to falsify the result before accepting it. Use when a MiniMax Code session, subagent or cloud agent reports a task done, before merging its PR or acting on its analysis.

1h installs
1
GitHub stars
685
Updated
Nov 10
minimax-ai
Community

handoff-doc

Write a hand-off document so a fresh session or another person can continue the current task without the chat history. Use when switching tasks, after several compactions, before clearing a session, or when asked to write a handoff, brief, or progress doc.

1h installs
1
GitHub stars
685
Updated
Nov 10
minimax-ai
Community

lifting-loops

Finds loops where the user keeps acting as a manual step for agents (retrying, judging output, approving, carrying context, starting the same task by hand, doing the agent's work) and proposes how to lift the user out: delete a stale rule, change a skill / AGENTS.md / playbook, add a script or metric, add a skill or scheduled task, or advise the user. Keeps a private loop ledger and promotes Claude Code auto-memory into artifacts. With a task already in the session, reflects on that session; in

1h installs
1
GitHub stars
685
Updated
Nov 10