Pimp My IDE / garage dispatch
Back to garage
October 2, 2026 | agent harnesses / release design

Minimal is a release rule.

Pi 1.0 adds Codemode, deferred tool loading, virtual models, and a fullscreen terminal interface. The stronger idea is the filter that kept other features out.

Make every harness feature pay for its prompt cost, failure path, and removal cost before it becomes a default.

The artifact clears the first gate.

Pi 1.0 shipped as an MIT-licensed release with source archives, platform binaries, checksums, and install manifests. The Linux x64 archive passed its published SHA-256 check in a disposable directory. Its binary reported version 1.0.0 and printed the documented command and option routes.[2]

That proves the release artifact exists and starts on one Linux host. It does not prove provider setup, model quality, extension compatibility, or production fit.

The feature list is really a loading policy.

Earendil says Pi waits for features to prove useful before adopting them. Version 1.0 adds Codemode, virtual-model extension support, deferred tool loading, Anthropic cache warming, mid-conversation system messages, a new theme, and fullscreen mode.[1]

The release notes make one tradeoff concrete. With the default tools and Codemode active, a GPT-5.6 request dropped from about 5,300 prompt tokens to about 3,300. Tool descriptions became short pointers, while detailed reference material moved behind demand.[2]

A small default is not a lack of ambition. It is a demand that each new path name its cost.

Do not confuse optional with free.

Codemode can call MCP tools and image models. Pi 1.0 also hardens MCP OAuth checks, stores credentials by server name and URL, and preserves granted scopes during step-up sign-in. Those are useful changes. They also create more states to test, document, and recover.[2]

Deferred loading reduces prompt weight. It does not remove runtime authority. An unloaded tool may still become available later. Keep the permission boundary, credential scope, and audit record separate from the prompt-size decision.

Give every addition an admission card.

  1. Name the task that cannot be handled by the current default.
  2. Measure the prompt and latency change on that task.
  3. Exercise one missing-tool or failed-authentication route.
  4. Keep a setting, extension boundary, or package split that can remove the feature.
  5. Promote it only after the same workload improves without hiding new failure states.

The governor below drafts that admission card. It does not install Pi, configure a provider, enable a tool, or validate a workload.

Interactive makeover / feature governor

Set the admission load.

Traditional purpose replaced: one feature checklist. Better version: a native workload selector and independent admission gates drive one mechanical shaft, one verdict, and a copyable review card.

Choose the job

These workloads are an editorial review pattern. They are not Pi modes.

Target workload
Admission card sections
Admission shaft

Terminal edit loop

0 of 4 sections selected
TaskCostFailureRemoval

Feature case is open.

Name the task gap before adding weight to the default terminal loop.

This bench selects review-card sections. It does not supply measurements, approve a feature, change Pi settings, or prove that a tool is safe.

Feature admission card

The all-selected state means the card structure is complete. Real measurements, revisions, and test output remain required.

Sources read

Source log and evidence boundary
  1. Earendil, "Pi 1.0", read October 2, 2026. This supplies the release framing, feature list, minimalism policy, Pi Durable split, install routes, and license claim.
  2. Pi v1.0.0 release, read October 2, 2026. This supplies the release assets, commit, Codemode token comparison, fullscreen default, MCP OAuth changes, deferred-tool repair, and other release details. Pimp My IDE downloaded the Linux x64 archive and SHA256SUMS, verified the archive, then ran pi --version and pi --help in a disposable directory.
  3. Pi source repository, read October 2, 2026. This verifies the public monorepo, package layout, documentation, examples, and MIT license.
  4. Hacker News discussion for Pi 1.0, item ID verified through the Hacker News API and read October 2, 2026. This is the discovery and practitioner-discussion route, not technical authority.

Evidence boundary. Pimp My IDE verified one release archive and the binary's version and help output. We did not configure a provider, run a model request, start MCP, load an extension, reproduce the token comparison, or test Pi on macOS or Windows. The feature governor is an editorial review pattern, not a Pi feature.