How to Auto-Ship Linear Issues as PRs with OpenAI Codex

OpenAI Symphony Codex Linear PR Orchestrator polls Linear for unassigned issues, assigns Codex agents to claim and implement each one, and creates a pull request — all without human intervention. Issues with clear acceptance criteria ship in 30-90 minutes instead of waiting for developer assignment. Teams save 30-50 hours weekly on triage, context-switching, and manual implementation.

A team of 8 engineers receives 25-40 Linear issues per week. The tech lead spends 6-8 hours triaging and assigning these. After assignment, each developer spends 30-60 minutes per issue understanding context and setting up a branch.

[ STAT ] Knowledge workers spend 19% of their workweek searching for and gathering information. For engineering teams, that translates to 7-8 hours per developer per week lost to context acquisition. — McKinsey Digital, 2024

The cost compounds. A team of 8 at $150/hr blended rate loses $4,000-6,000 per week to overhead that does not produce any code.

[TOOL: OpenAI Codex] Autonomous agent that claims issues, analyzes codebase context, implements solutions, writes tests, and creates PRs. Each issue gets a dedicated ephemeral workspace. [TOOL: Linear] Issue tracking system. The Symphony polls Linear's API, updates issue status, and posts architecture analysis comments.

The agentic reasoning step happens during clarity scoring. Codex reads each issue's title, description, and comments, then scores clarity on a 1-10 scale. Issues below 6 get a needs-more-detail label. This decision determines whether the issue enters the autonomous pipeline or waits for human clarification.

For engineering teams of 6-15 developers using Linear for sprint management: your tech lead currently spends 6-8 hours per week on issue assignment. This workflow eliminates the assignment step entirely for well-specified issues.

For startup CTOs acting as sole architect and reviewer: you read every issue and assign each one personally. Codex handles 40-50% of issues autonomously, leaving only complex architectural decisions for your review.

For open-source maintainers managing 100+ open issues: Codex per issue lets you ship fixes for labeled bugs and simple features without diverting core maintainer time.

Issue Polling. Symphony polls Linear every 60 seconds for unassigned issues matching configured labels.
Clarity Scoring. Codex scores each issue on clarity using a rubric: acceptance criteria presence, error reproduction steps, expected output format. Issues below 6 are skipped.
Workspace Creation. Codex creates a Git branch, virtual environment, and workspace directory. Issue marked in-progress in Linear.
Architecture Analysis. Codex reads relevant codebase modules and writes an implementation plan as a Linear comment.
Implementation. Codex writes code, tests, and config changes. Runs existing tests to confirm no regressions.
PR Creation. If tests pass, Codex commits, pushes, and creates a PR with issue number and summary.
Human Review. Developer reviews the PR. Changes requested trigger iteration in the same workspace. Approval triggers merge and workspace destruction.

60 minutes. That is the honest setup time. Most of it is getting the Codex API wrapper to handle Git operations correctly.

OpenAI Codex (API, GPT-4o) → Autonomous agent. API key from platform.openai.com. Linear (GraphQL API) → Issue tracking. API key from linear.app settings.

Gotcha: Codex does not have built-in Git operations. You must wrap the API calls in a Python script that handles branch creation, commit, and PR via GitHub CLI. The wrapper script is 200-300 lines of Python and accounts for most of the 60-minute setup time.

▸ Issue triage time: 6-8 hrs/week for tech lead → under 30 minutes for exception review ▸ Developer context-switch per issue: 30-60 minutes → 5-10 seconds reading architecture comment ▸ Issues shipped per week: 15-20 manually → 30-40 with autonomous agents ▸ First-week measurable: percentage of new issues claimed by Codex within 5 minutes — target 40%+ ▸ Cost per shipped issue: $150-300 in developer time → $1-5 in API costs

Codex can only ship issues with clear acceptance criteria. Ambiguous issues get tagged needs-more-detail and require human rewriting before entering the pipeline.
A complex issue requiring 200K+ input tokens costs $3-8 per attempt. If the first implementation fails tests, cost multiplies. Set a per-issue budget cap.
Each issue creates a long-running branch. If the issue stays open for days, merge conflicts accumulate. Configure the workspace to rebase daily on main.
(5 min) Get your OpenAI API key at platform.openai.com. Verify codex model access.
(15 min) Get your Linear API key at linear.app/settings/api. Note the team ID and label IDs for filtering.
(30 min) Write or clone the Codex agent wrapper script. This is the longest step — the script must handle Git operations, Codex API calls, and Linear status updates.
(10 min) Deploy Symphony as a background process or cron job. Point it at a test Linear team and a staging repo. Watch it claim and ship its first issue.

Q: How many issues can Codex handle per day? A: With standard API tier access, Codex can process 30-50 well-specified issues per day. The bottleneck is total token throughput — each issue consumes 50K-200K input tokens for codebase context plus generation.

Q: Does Codex understand my existing codebase conventions? A: Codex reads relevant module files during the architecture analysis step. It infers naming conventions, file structure, and patterns from the code it reads. Providing a CLAUDE.md or similar style guide in the repo root improves consistency.

Q: What happens if Codex creates a PR with failing tests? A: The PR is still created but marked as draft. Codex posts the test failure details in the PR description. A developer must investigate and fix the issue before the PR can merge.

Q: Can I limit which issues Codex can claim? A: Yes. Configure the polling filter by label, priority level, or project. Most teams start with label:bug and label:small-feature before expanding scope.

Q: What is the risk of Codex making incorrect architectural decisions? A: Codex reads only the relevant codebase modules, not the full system architecture. For changes affecting cross-cutting concerns (auth, data layer), the architecture analysis may miss dependencies. Use labels to exclude cross-cutting issues from the autonomous pipeline.