Skip to content

Releases

The Effort Project 0.3.0 — Experimental

Choose a tag to compare

@AntyXO AntyXO released this 18 Nov 21:12
e07bc5c

Third experimental release: the Claude Code plugin now suggests a model and effort on every prompt and routes only when asked, Claude's daily tier is Sonnet, the rules cover more words and languages, and model routing has its first benchmark. Equal quality at lower cost is still not shown.

Added

  • Benchmark arms that route the model as well as the effort (--arms routing), six harder tasks, and the first model-routing results in bench/PILOT.md. On Claude, routing cost about a sixth of always using the frontier model and no less than defaulting to the daily model.

Changed

  • The Claude Code plugin now shows a model and effort suggestion for every eligible prompt without changing Claude's behavior. Set EFFORT_AUTO_ROUTE=1 in Claude Code's settings.json env to opt into the existing low/high skill routes; medium stays advisory. Short follow-ups without a rule signal can inherit a recommendation from up to three earlier user prompts in the last 64 KiB of the local transcript, read without storing or logging its text.
  • Claude's built-in daily model is now sonnet instead of opus; fast remains haiku and frontier remains fable.
  • Policy rules-0.3 recognizes more costly-error, investigative and mechanical English keywords, including login, passwords, billing, destructive data changes, privacy, crashes, performance, comments, lint and version bumps. Basic Spanish, French, German, Portuguese and Italian keyword coverage uses Unicode word boundaries. Risky and investigative words still take precedence over short mechanical edits.
  • Policy rules-0.4 also treats money movement as risky: funds, balances, ledgers, refunds, payouts, withdrawals, deposits and money or bank transfers, with basic Spanish, French, German, Portuguese and Italian equivalents. Such prompts start at high effort; the model tier is unchanged unless the prompt also asks for investigation. Added after every benchmark failure turned out to be the node-bank-ledger task, so that task no longer counts as held out for this rule.

Fixed

  • effort auto codex no longer drops the official terminal UI's session when Codex sends a single message larger than 8 MiB, which happens right after a turn starts for users with many skills, plugins or MCP servers. The router's message limit is now 64 MiB.
  • Two tests added in 0.2.0 now pass on Windows. The route-skill test accepts the CRLF line endings that Git for Windows checks out by default, and the WebSocket terminate test waits for the server end to finish closing, not only the client end. This changes test code only.
  • On Windows, command lookup now finds programs installed as app execution aliases, such as the py launcher that the Python install manager puts in %LOCALAPPDATA%\Microsoft\WindowsApps. Node cannot read these entries, so a verification command such as py -m pytest stopped with "was not found on PATH". The benchmark harness now picks only a Python that this lookup finds, and its checks run the interpreter it probed.

Install

npm install --global github:AntyLab/effort-project

Claude Code plugin: /plugin marketplace add AntyLab/effort-project, then /plugin install effort-project@effort-project.

The Effort Project 0.2.0 — Experimental

Choose a tag to compare

@AntyXO AntyXO released this 16 Nov 18:05
2f97d2e

The second experimental release of The Effort Project. Effort now picks the model as well as the effort level, for Claude, Codex and Gemini, so you do not have to choose either. Still MIT licensed, with no npm dependencies, telemetry, or install hooks.

What is new

  • Model and effort together. Short mechanical requests go to a fast model at low effort, requests that are both high-stakes and tangled go to a frontier model at high effort, and everything else stays on the daily model. Built-in models: Claude haiku / opus / fable, Codex gpt-6-luna / gpt-6.1-sol / gpt-6-astra, Gemini gemini-4-hydrogen / gemini-4-argon. --model pins a model; effort auto --routing-config overrides the table.
  • Claude Code plugin that routes by itself. Its prompt hook asks Claude to invoke a routing skill: Haiku at low effort in a fork, Fable at high effort in a fork, or high effort on the session's model. Ordinary prompts are left alone.
  • Codex terminal router. effort auto codex now starts with a small built-in WebSocket server and needs no extra package.
  • Gemini. effort run --provider gemini and effort auto gemini, through the Antigravity CLI (agy). File permissions last for the session only and shell commands are never allowed.
  • Benchmark harness with eleven small tasks, offline by default, and a live acceptance script.
  • The npm package is now named effort-project (previously @antyxo/effort-project).

The full list, including fixes, is in the changelog.

Install

Requires Node.js 22+. Download effort-project-0.2.0.tgz below, then run:

npm install --global ./effort-project-0.2.0.tgz
effort doctor
effort recommend "Investigate an intermittent race condition" --provider codex

If you installed 0.1.0 globally, remove it first with npm uninstall --global @antyxo/effort-project. Managed tasks need an installed, signed-in Codex, Claude Code or Antigravity CLI; your existing subscription or API usage still applies. The Claude Code plugin and source installation are described in the README. SHA256SUMS contains the download checksum.

Validation and limits

Checked live on macOS on 2026-11-16 (Node 26.13, Claude Code 2.1.310, codex-cli 0.160.0, agy 1.5.9):

  • effort run on a typo, a small feature and a race condition chose the expected model and effort and passed its check on Codex and Gemini (3 of 3 each). On Claude, the daily and frontier runs passed, and the Haiku run passed 6 of 6 after a fix made in this release.
  • The Claude Code plugin, run headless from this repository, routed a typo to Haiku at low effort and a race condition to Fable at high effort, and left an ordinary prompt unchanged.
  • The Codex router rewrote a typo prompt from gpt-6.1-sol at max effort to gpt-6-luna at low, and Codex's own session record agreed.
  • 55-run benchmark pilots on Codex (gpt-6.1-sol) and Gemini (gemini-4-argon): every arm passed every task, so the comparison with fixed high effort is inconclusive.

All 367 local tests passed (10 skipped), with the syntax check and the offline package installation check. Continuous integration is switched off for this repository, so there is no CI run for this release.

See the validation record for the full evidence.

This release uses uncalibrated rules. It has not established cost savings or equivalent coding quality. Frontier models cost more, and a run has no spending cap. Claude Code routing depends on Claude following the hook, and a forked skill does not see earlier conversation unless Claude passes it along. Desktop MCP tools still only give advice. Not checked: a prompt typed into the official Codex terminal UI, OpenCode, the plugin installed from GitHub rather than a local folder, and Windows or Linux live runs. Review the README and SECURITY.md before running tasks.

Contributions and reproducible compatibility reports are welcome. Do not post credentials, provider transcripts, or private task prompts in issues.

The Effort Project 0.1.0 — Experimental

Choose a tag to compare

@AntyXO AntyXO released this 26 Sep 12:58

The first public experimental release of The Effort Project: a free, local controller for reasoning effort in coding agents. MIT licensed, with no npm dependencies, telemetry, or install hooks.

What is included

  • Codex and Claude Code CLI adapters that request real effort levels and preserve sessions across managed attempts.
  • Transparent low/medium/high recommendations, explicit effort/attempt/time limits, independent verification, and bounded escalation after repeated test failures.
  • A local dashboard for recommendations and run history.
  • Advisory MCP tools for compatible desktop hosts, with generated setup entries.

Install

Requires Node.js 22+. Download antyxo-effort-project-0.1.0.tgz below, then run:

npm install --global ./antyxo-effort-project-0.1.0.tgz
effort doctor
effort recommend "Investigate an intermittent race condition" --provider codex
effort dashboard

Managed tasks also require an installed, signed-in Codex or Claude Code CLI. Existing provider subscriptions/API usage still apply. Source installation instructions are in the README. SHA256SUMS contains the download checksum.

Validation and limits

Live macOS checks completed low/high turns in the same session with both providers. Each provider also repaired a disposable module at low effort; independent positive/negative/zero tests passed and the test files were unchanged. Offline tests cover permission failures, session/model mismatches, recovery, process cancellation, usage accounting, MCP and dashboard access controls. A package smoke check installs the archive offline and exercises its CLI, MCP and dashboard.

All 70 local tests passed. The release CI run passed all six jobs: Linux, macOS, and Windows with Node 22 and 24, including the offline package installation check. POSIX-specific cleanup assertions are explicitly skipped on Windows. The first matrix found a real cancellation race; the release includes its verified fix.

See the validation record for the full evidence and limits.

This release uses uncalibrated rules. It has not established cost savings or equivalent coding quality. Effort changes happen at managed turn boundaries. Desktop MCP tools provide recommendations; they cannot change a host chat's active effort. Live provider compatibility was checked on macOS only. Explicit verification commands run with the user's normal local privileges. Review the README and SECURITY.md before running tasks.

Contributions and reproducible compatibility reports are welcome. Do not post credentials, provider transcripts, or private task prompts in issues.