mini-SWE-agent

Nothing is hidden, and uvx mini-swe-agent has it running before the coffee is done, which is the daily trait: a hundred lines you can read in full, so when it does something odd you know why by lunch. That makes it the best tool here for comparing models on the same task, because the harness contributes almost nothing.

You will not like it as a daily driver, because there is no IDE plugin, no browser and no comfort features, and every convenience is yours to add. Pick it for research and model comparisons. Pick Claude Code or Aider for the sprint, where the comfort features are the product.

The risk is that simplicity is a policy. The docs say that for tasks like opening PRs you tell the LM to figure it out, so the model improvises where other harnesses have a procedure, and improvisation is where the wrong git command lives. Without an environment flag, commands land on your machine.

The consequence: never run it outside a container on a repository you care about, and expect the model, not the harness, to decide how a task ends. A documented procedure for the common endings would change this verdict. What it does right: Docker, Podman, Singularity, Bubblewrap and Modal are all first-class sandboxes.

The cleanest methodology here. 1. One tool, bash, executed with subprocess.run, each action independent, so there is no hidden state between steps. 2. A linear history; every step appends, nothing is summarized, so the transcript is the full context. 3. No use of the model's tool-calling interface, which removes a vendor-specific variable. 4. The above 74% on SWE-bench Verified is reported with Gemini 3 Pro on the bash-only setup.

The consequence is that the score is attributable: change the model and nothing else moves. The observation: when the scaffold is this small, the score is the model's.

No company to evaluate, which is clarifying. The SWE-agent group at Princeton and Stanford maintains it, the star count is 6,938, and revenue is zero by design, funded by grants and paid in citations. The moat is academic: the group also runs the benchmark everyone cites, so the harness outlives any startup that scores on it.

The risk is graduation, since labs turn over and a reference harness needs a maintainer with a reason. No acquirer, no pivot, no price to raise. Position: no vendor risk because no vendor; use it as a yardstick, not a product.

The demo is a model solving a benchmark task in a terminal. Procurement has nothing to sign: no seat, no SSO, no SCIM, no retention terms, and Windows is not listed, so a fifth of the fleet cannot run it. It runs headless, so CI could use it as an evaluation step when we compare models before a contract renewal.

Onboarding is a pip install and a config file, which is also the whole support plan. Approved with conditions: the evaluation team only, never the fleet, and a container policy written before the first run.

Full ownership. MIT license, any model through litellm, OpenRouter or Portkey, so my local box is a model string away, plus a YAML config for the prompt and the environment. The agent is short enough to read in full, and a fork is a copy, which is the only definition of forkable that has ever held.

No MCP client, an afternoon's work, and the kind of afternoon I enjoy. Sandboxes are a flag. This is the tool I would write if I had a benchmark to defend, and the fact that the benchmark's authors wrote it is the point.