多代理組團把百道難題多解十二題

FilesExpand file tree77f4003de7517f016331ac5bbdbdc0ffe3f00fd1/README.mdCopy pathBlameMore file actionsBlameMore file actions Latest commit HistoryHistoryHistory113 lines (92 loc) · 5.48 KB77f4003/README.mdCopy pathTopFile metadata and controlsPreviewCodeBlame113 lines (92 loc) · 5.

48 KBRawCopy raw fileDownload raw fileOutlineEdit and raw actionsteam-solve A multi-agent harness for programming problems, written as a Memdoor workflow.Five roles (planner, two ideators, solver, manager) and a repair worker each start from a fresh context and talk only through files on disk.

A shell script that runs the samples decides success, not a model.It is the method of GVS5H (Lee, 2026: five Qwen3.8-27B models match Fable 5 on LiveCodeBench Hard).On GVS5H's 100 hardest LiveCodeBench problems, with Qwen3.

8-27B and thinking off, the team solved 47 against 35 for a single call of the same model, for $3.27 in all (bench/ has the harness, every problem's result and what the numbers do not show).The coder wrote these files itself from a description of the method (2026-10-25).Nobody hand-wrote them.

Run it curl -fsSL https://memdoor.ai/install.sh | bash memdoor connect openrouter # or anthropic, openai, gemini: your own key git clone --depth 1 https://github.com/guregodevo/memdoor mkdir -p ~/.memdoor/workflows cp -r memdoor/examples/workflows/team-solve ~/.

memdoor/workflows/ # the library: any project sees it mkdir my-problem && cd my-problem cp -r ../memdoor/examples/workflows/team-solve/sample/* .# task.md + tests/1.in … 5.

out memdoor workflow run team-solve --wait # progress on stderr, exit 0 when every sample passes /workflow in memdoor tui draws the same run live.For your own problem, replace task.md and put paired tests/N.in / tests/N.out files beside it.The program is solution.

py, read from stdin, compared to the samples with diff.It needs Python 3 and bash.To keep it with one repo instead, copy the folder into that project's .memdoor/workflows/.The project's copy wins.The steps name no model.

Whatever /model is pinned to answers, a local one included (memdoor connect local for janis, Ollama or llama-server).How it works plan ──┬── ideate-a ──┬── solve ── test ── review ── repair ── verify └── ideate-b ──┘ step kind reads writes plan agent task.md plan.md, and a notes.

md ledger with one anchor per ideator ideate-a, ideate-b agent, in parallel task.md, plan.md, notes.md a proposal at its own anchor: A the most direct provable algorithm, B one from a different family of technique solve agent both proposals the decision in notes.md, solution.

py test command tests/ test-output.log, test-status.txt (never fails the run) review agent the captured test output, not a model's claim review.md, repair instructions in notes.md repair agent the manager's instructions a fixed solution.

py, or nothing if the samples passed verify command tests/ final-test-output.log; nonzero exit fails the run Every agent prompt says "fresh context, read the files, do not rely on chat".The paper credits that ledger discipline for the gain.There is one repair round and no human gate.

The last step is the whole verdict: type: command requires: - tablepattern: repair command: | check=.memdoor/workflows/team-solve/check-samples.sh [ -f "$check" ] || check="$HOME/.memdoor/workflows/team-solve/check-samples.sh" bash "$check" > final-test-output.log 2>&1 result=$?

cat final-test-output.log exit "$result" Each step is one YAML file in steps/, the format of mario, the engine underneath.requires: is the DAG.Changing a role means changing its prompt.

What the first run found The sample: count the substrings of a string over {a, b, c} with equal counts of each letter, up to 200,000 characters.Run on the default ladder through OpenRouter: 8 of 8 steps, 5 of 5 samples.

The solver took ideator A's prefix-state method, O(N) with a dictionary of (a−b, a−c) pairs.A brute force on 300 random strings found no mismatch afterwards, and it runs in 0.04 s at N = 200,000.The manager found nothing to repair.One ideator's provider dropped the stream mid-answer.

The coder had written maxretries: 0, so the run died there.The agent steps now carry max_retries: 1, and the run was resumed with --partition, which kept the planner's and the other ideator's work.

The ideator prompts the coder wrote named the sample's methods (A: a streaming prefix-state count, B: sorting prefix states), which would steer any other problem the wrong way.They now ask A for the most direct provable algorithm and B for one from a different family of technique (2026-10-28).

The first attempts failed every step in 0 s with "no agent named coder": a run looked its agent up only in its own workspace.That was fixed in the gateway the same day (v1.692.0).

On the benchmark bench/ runs the same method with every role as one model call (team-llm-lcb) against a single call (single-lcb).

Both are Memdoor workflows on the same model and host, graded on every public and hidden test with LiveCodeBench's own checker: arm solved AtCoder (73) LeetCode (27) single call 35 / 100 32.9% 40.7% team workflow 47 / 100 42.5% 59.3% These are thinking-off numbers, not the paper's setting (66.

8% → 92.4% with 128K tokens of thinking).The finding is the gap between the two arms.The bench README says why, and how to run it again for about four dollars.