v0.4.4

Patch Changes

• dafbe67: Update now works where the desktop app puts no claude on the PATH: the mod looks for Claude Code through a shell, and if it still cannot run it, asks Claude to run the update in the chat, and gives a terminal command in case that is refused.

v0.4.3

Patch Changes

• 6145ac9: Update now also sits beside Check now in the pane's Updates section, so a check that finds a new version needs no scrolling. When Claude Code cannot be run from the pane, the pane shows the real error, and the command to run names Claude Code by its full path.

v0.4.2

Patch Changes

• 872716c: A new version is now plain to see: the band shows a highlighted Update button that opens the pane, in place of the dot on ≡, and Reload once updated. The update card has more room, and the band and the card wrap on narrow screens. Sessions already open now learn of a new version by themselves, without pressing Check: each looks every 30 minutes at what the machine read, and GitHub is still read at most once a day per machine, so a release can take up to about a day to show.

v0.4.1

Patch Changes

• af0144c: The mod's notes to Claude no longer add messages to the conversation, which broke the prompt cache. They now go with your prompt or with a tool's result. A new request waits for Haiku's reading, up to 15 seconds, so its first request goes out at the right effort and Claude asks about the flow or a stronger model before it starts. The band says when Haiku is reading.

v0.4.0

Minor Changes

• 537263b: open-effort now tells you when a new version is out. Once a day (switch it off in the pane) and whenever you press Check now, it reads the newest version number from GitHub; when one is out, the band's ≡ shows a dot and the top of the pane shows Update now, which runs Claude Code's own plugin update, then Reload now. Where the update cannot run, the pane gives the command.

v0.3.0

Minor Changes

• f905caa: Climbing past the top effort no longer jumps to Fable 5.1 by itself. Claude stops and asks before moving the task up one model (Sonnet → Opus → Fable 5.1), and with a yes the task carries on there from high. Keeping the model means three more failing checks stop the work. A ticket's sub-agent climbs no further than Opus 5.5.

v0.2.0

Your own flow

A project can now use its own flow instead of Matt Pocock's Main Flow. Put it in .claude/open-effort.json, or type /open-effort flow and Claude asks you about your flow and writes the file once you agree.

• Name your stages (2 to 10), the skills that run each, and the effort each starts at.
• Optional tickets: the stage that writes them, the stage that works them, and where their files live. Each ticket keeps its own Model: and Effort:.
• No file: Matt Pocock's Main Flow works as before.
• A file with a problem keeps the flow off, and the pane says what is wrong.

Also

• The pane's Main Flow section is now Flow, and names the flow the project uses.
• Meets the plugin directory's checks.
• The README shows three ways to install, with a picture, a Try it section, and what each hook does and sends where.

Install

claude plugin marketplace add JakkapatB/open-effort && claude plugin install open-effort@open-effort

Already installed? Update it:

claude plugin marketplace update open-effort && claude plugin update open-effort@open-effort

Full list: CHANGELOG.md

v0.1.0

First release of open-effort: a Claude Code mod that fits Claude's effort and model to the work.

• Haiku reads each new request and sets a starting effort; three failing checks in a row raise it one level, then move to Fable 5.1, then Claude stops and asks you.
• Asks before moving a task to a more capable model.
• Matt Pocock's Main Flow: an effort per stage, tickets with their own Model: and Effort:, a sub-agent per ticket.
• Sub-agents by type: Explore on Haiku at low, Plan and general-purpose on the task's model.
• A band above the prompt, and a pane with switches, the Main Flow's model, language, four themes and a history chart.
• English and Thai.

Measured over 61 sessions on Opus 5.5: 4–11% lower cost on coding tasks with the same results, and 78% lower cost for Explore sub-agents. See the README.

Install: /plugin install open-effort --marketplace JakkapatB/open-effort