Spend leftover AI quota on reviewable PRs
Hey folks; I’m the author of RemainderBot. Task: The author, a startup founder, used RemainderBot to automate small, context-heavy tasks by leveraging unused Claude and Codex subscription quotas. The tool reads personal goals and task lists, selects a task, and generates a GitHub Pull Request (PR) containing the completed work. The human reviews the PR, comments on it, and decides whether to merge, which guides the bot's future task selection and approach. The bot can run manually or as a cronjob, checking remaining quota before executing.; tools: Claude, Codex. Human role: The original post does not say how a person stayed involved.; limitations: The tool requires authenticated git, claude/codex, and gh CLIs on the user's machine. It depends on the user maintaining accurate GOALS.md and sources.toml files. The bot's effectiveness relies on the quality of human feedback via PR comments to learn preferred task types. Running as a cronjob requires a server environment. The source does not detail specific failure modes or costs beyond 'a few percent of weekly quota' per task..
Tools
Published on Agent Work Atlas:
Overview
Hey folks; I’m the author of RemainderBot.
Workflow
The author, a startup founder, used RemainderBot to automate small, context-heavy tasks by leveraging unused Claude and Codex subscription quotas. The tool reads personal goals and task lists, selects a task, and generates a GitHub Pull Request (PR) containing the completed work. The human reviews the PR, comments on it, and decides whether to merge, which guides the bot's future task selection and approach. The bot can run manually or as a cronjob, checking remaining quota before executing.
Reported result
The author reports that RemainderBot successfully created 111 PRs, including customer call preparations, a census of agent setup files across 31 GitHub organizations, and standalone prototypes for demos. The system effectively utilized leftover subscription tokens that would otherwise have been wasted. No specific metrics on time saved or error rates are provided, but the volume of completed PRs indicates consistent functionality.
Human role
The original post does not say how a person stayed involved.
Limitations
The tool requires authenticated git, claude/codex, and gh CLIs on the user's machine. It depends on the user maintaining accurate GOALS.md and sources.toml files. The bot's effectiveness relies on the quality of human feedback via PR comments to learn preferred task types. Running as a cronjob requires a server environment. The source does not detail specific failure modes or costs beyond 'a few percent of weekly quota' per task.
Try it
I want to set up RemainderBot to automate small tasks using my leftover AI subscription quota. First, verify that git, claude (or codex), and gh CLIs are installed and authenticated on my machine. Clone the openremainderbot repo from https://github.com/arimwilson/openremainderbot.git. Copy GOALS.example.md to GOALS.md and sources.example.toml to sources.toml. I will provide my personal goals and task list sources for you to edit into these files. Once configured, run 'python3 -m remainderbot snapshot' followed by 'python3 -m remainderbot run --provider claude --deadline-minutes 20 --size small'. This should pick a task, create a branch, and open a PR. Please show me the PR link and the task description before I review it. If the task is incorrect, I will provide feedback via PR comments. Do not merge any PRs without my explicit approval. Start with a single small task to test the workflow.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.