ProPR 0.9.2 · 18 November 2026

PR status, Automation templates and Claude Fable 5.2.

Open the pull request list and read which ones need you. Start an automation from a template or from a blocked epic issue. Pick Claude Fable 5.2, GPT-6.1 Terra or Gemini 4 Hydrogen, run an npm package as a local MCP server, and export what every model costs.

Overview

New

Every pull request says whether it needs you.

ProPR gives each pull request it opens one status, Working, Needs you or Ready, and shows it where you already look: the first line of the description and the start of the title.

One live status line

The top of every open ProPR pull request reads, for example, ✅ Ready · review 8/10 · checks: 35 passed. It names the running activity or Ultrafix cycle, the latest review score, the check counts and the reason a PR needs you. A PR counts as ready only when its checks passed on the current head and GitHub confirms it does not conflict with its base.

The state in the title

The title starts with 🔨 while ProPR works, ⚠️ when the PR needs you and ✅ when it is ready, the last two followed by the review score as a solid digit (✅❽). The pull request list reads at a glance, and merge commits, task titles and notifications use the title without the slot. The prefix is a per-repository setting, on by default.

Epic progress where you look

An epic pull request counts its merged issues in its title (◑ 3/5 [Epic] Scheduler primitives) and its status line (3 of 5 issues merged · #3475 in progress), and turns ✅ 5/5 when the last one lands. Its progress comment lists what is in progress and what is up next.

The worker brings every open ProPR pull request up to date when it starts, including ones opened before this release, and removes the status when a PR merges or closes. MCP's get_pull_request, list_pull_requests and get_work_overview return the same status. What each status means · The epic playbook

New

Automations: start from a template, or from a blocked epic issue.

Three templates fill in the brief, capabilities, schedule and previous reports; you pick the repositories and adjust the autonomy. A new event trigger runs an automation the moment an epic issue gets blocked.

  • Daily issue triage

    A report of the issues that need a human, likely duplicates, proposed severity and security labels, and stale issues.
  • Instruction learning loop

    A weekly read of recent pull request discussions, including GitHub reviews and inline review comments, that proposes short additions to or removals from the repository's agent instructions. Each candidate cites at least two pull requests, and with autonomy raised the acting step files one task that opens the pull request.
  • Weekly codebase drift report

    The last seven days of merged changes by area, new code that duplicates existing helpers, growth per area and suggested consolidations, which the acting step saves as repository To-Dos.
  • Run on a blocked epic issue

    Subscribe an automation to the plan.issue_blocked event and it runs once per blocked issue, with the issue, plan and blocker reason as its input. The trigger is opt-in, limited to the automation's repositories and optionally to some reasons, and runs at most three times for the same issue.
  • Runs in the task list

    An automation run's row is labelled Automation and carries the automation's name.
  • Repeat without cron

    The schedule is Off or Repeat: every few hours, daily or weekly at a time you pick. A cron expression stays available behind an advanced link.
  • Find an automation

    The Automations list has the same header as the other screens, with a repository filter and a search box.

The template catalog · Triggers · Automations guide

An issue triage report: one payments issue to act on first, two likely duplicates, two issues missing reproduction steps
The Daily issue triage template's report: what to act on first, likely duplicates, and issues missing reproduction steps.

Updated

Claude Fable 5.2, GPT-6.1 Terra and Gemini 4 Hydrogen.

Claude Code

Claude Fable 5.2 is the new target of the plain fable alias and of the recommended plan-generation and PR-review models. You still pick Claude Fable 5.1 as a legacy model through fable51 and llm-claude-fable51.

Claude Code in ProPR

Codex

GPT-6.2 Astra is the new target of the astra alias and llm-codex-astra, and GPT-6.2 Sol (llm-codex-gpt62-sol) is the recommended Sol; GPT-6 Astra and GPT-6.1 Sol stay selectable as legacy models. GPT-6.1 Terra joins as llm-codex-gpt61-terra, with reviews on the 500K Codex runtime window.

Codex in ProPR

Antigravity CLI

Gemini 4 Hydrogen, Google's fast Gemini 4 model, arrives as llm-antigravity-hydrogen with the hydrogen alias. It is the only Antigravity model with xhigh effort.

Antigravity CLI in ProPR

Claude Sonnet 4.5 retires

Anthropic retires Claude Sonnet 4.5 on 30 November 2026. It is selectable until then; from that date, runs launched with a saved Sonnet 4.5 selection, alias or label use Claude Sonnet 5.5. Opus 4.5 and Haiku 4.5 are unaffected.

Pinned CLIs

The propr/agent image moves to Claude Code 2.1.326, Codex 0.169.0, Antigravity CLI 1.5.10 and OpenCode 1.19.6, with Mistral Vibe 2.25.8 as in 0.9.1. Claude Code's new permission prompts never reach ProPR runs.

Timeouts, explained

An Antigravity run cut short by its print timeout is recorded as timed out with the CLI's own explanation, and analysis runs no longer accept the partial answer.

New

Local MCP servers from npm, bigger plans, and goals that ask.

One extension library

Extension profiles give way to one list of MCP servers and skills, each limited to the repositories and agents you choose. Enabled items apply automatically to matching task runs, pull request follow-ups and automation runs; the disable-mcps, disable-skills and disable-extensions labels switch them off for one issue or pull request. An HTTPS server can sign in with OAuth from Settings.

Agent extensions

Local MCP servers from npm packages

An administrator adds an npm package at an exact version, with its arguments, variables and write-only secrets, and ProPR pins its integrity value. The worker installs the package once into its own cache, outside every task workspace, and each run mounts it read-only, checks the files against the pinned value and starts it for Claude Code and Codex inside the task container. Secret variables are readable there, so saving a server that has them asks for an acknowledgement. Python packages are not supported yet.

How agent extensions work

A Claude goal asks you a question

Claude goal sessions get an ask_operator tool: one question, optionally with suggested answers, shown under Needs you. Your next goal input is the answer. Pause and cancel work while it waits, and after 24 hours by default Claude is told nobody answered. Ordinary task runs never get the tool.

How Goals work

MCP plans hold 50 issues

create_plan, update_plan and publish_plan accept up to 50 plan issues, and implement_plan starts up to 50 per call, the same size Planner Studio always allowed.

MCP from chat

Reviews in the PR discussion

get_pull_request_discussion pages a pull request's submitted GitHub reviews and its inline review comments, with file, line and reply thread, next to the conversation it already read.

New

Spend by model and effort, and queues that respect capacity.

Cost export

Export recorded agent executions as JSON or CSV, grouped by day, repository, agent, model, reasoning level, execution type or outcome, with tasks, tokens and cost in USD per row. propr stats cost and the MCP tool get_cost_breakdown return the same data under the Analytics period and repository filters.

Cost in ProPR

Review quality by effort

Analytics' Agent efficacy by model shows each reasoning level under its model, the review-scores CSV has one row per model and effort, and propr stats review-scores prints merged PRs, runs to merge and cost per merged PR for each.

Scores by model

Auto-merge like a merge queue

Set auto_merge.require_up_to_date in .propr/workflow.yml and a labelled PR that passes every other gate but is behind its base is updated from it first, then merges once CI passes on the updated head. One PR per repository updates at a time, with at most three updates per PR and hour by default.

Ship an epic with auto-merge

Jobs wait off the worker

With a per-agent ceiling set, a job on a saturated agent waits in the queue with one Waiting for Agent Capacity row on its timeline, consumes no retry and no replacement, and starts oldest first when the agent frees up. Synthetic agents pass over a saturated member for another in the same tier.

Fixed

Fixes worth knowing.

Connected apps shows a full page

The Connected apps page counted revoked and expired grants towards each page of 50, so an account that reconnected often saw one or two apps per page. Pages now hold 50 active apps.

Consent shows that it is working

After Allow selected access or Deny, the MCP consent page shows a progress message, relabels the button and ignores a second click.

MCP-created plans open at Review

A plan created or updated over MCP with its issues now opens in Planner Studio at Review with the issues shown. Prompt-only drafts still open at Define, and a migration moves existing drafts that hold issues.

The summary is the summary

When an agent's last message only reacts to a late notification, ProPR's summarization model picks the message that describes the work for the completion comment and the commit subject. Without that model configured, nothing changes.

One CI comment per commit

Further failing checks on the same commit are added to the first CI-failure comment instead of posting a new one each, and start no second follow-up run.

Text stays selected

A finished run's report no longer re-renders every few seconds, so a selection in it holds.

A shorter link to the original

A visual preview whose full-size original is kept in ProPR Connect shows one Full-size original link under the image.

Upgrading to 0.9.2.

Back up the database and .env, then update the CLI and restart the stack. The CLI pins every service and agent image to its release.

npm install -g [email protected]
propr start --restart

Launcher installs pull the new propr/launcher image and restart. The maintenance guide covers backups, version pinning and rollback.

What changes on its own, and what waits for you

  • On from the first start

    The status line appears on every open ProPR pull request once the worker starts, and the title prefix is on per repository. Turn the prefix off under Repositories → Automation or with propr repo toggle owner/repo --pr-title-status off.
  • Your extension profiles

    Each server and skill from a profile becomes a library item with the same repository and agent limits and with Apply automatically off, and automations that used a profile keep the same items. No run changes until you turn an item on.
  • Opt in per automation

    The blocked-epic trigger is off until you set it on an automation. Templates fill a new automation as a dry run, so nothing acts until you raise the autonomy.
  • Off until you set them

    Local package MCP servers are added by an administrator in Settings. auto_merge.require_up_to_date defaults to false, and per-agent ceilings apply only with a per-agent policy set.
  • Before 30 November

    Move saved Claude Sonnet 4.5 selections to another Claude model, or let them run Sonnet 5.5 from that date.