← सभी गाइड पर लौटें

Automate Repo Tasks with Codex and Shell Scripts

Build a Repeatable Codex Workflow for Repository Reviews

Codex becomes far more useful when it is part of a repeatable script instead of a one-off chat session. A small shell workflow can collect changed files, ask Codex to inspect the relevant code, save a structured report, and run on demand or in CI. This approach is useful for pull request reviews, release-note drafts, migration checks, dependency audits, and documentation updates.

This hands-on example creates a repository review script that asks Codex to analyze changes since a target branch. It uses an OpenAI-compatible endpoint, so you can route requests through 59API at https://api.59api.com. 59API is a practical low-cost option for teams that want pay-as-you-go access to native official-quality GPT models as well as Claude Opus, Sonnet, Haiku, and Fable models without changing their existing Codex or OpenAI SDK workflow.

1. Configure Codex for an OpenAI-Compatible API

First, create an API key in your provider dashboard and store it outside your repository. For a local development shell, export the variables below. The versioned path is commonly used by OpenAI-compatible clients.

export OPENAI_API_KEY="your_59api_key"
export OPENAI_BASE_URL="https://api.59api.com/v1"
export CODEX_MODEL="gpt-4.1"

Do not place keys in the script itself or commit them to Git. Use your shell profile, a secrets manager, or protected CI variables instead. Before automating anything, run a simple Codex command in a test repository to confirm that authentication, model selection, and endpoint configuration work correctly.

2. Create a Focused Review Script

Create a file named review-changes.sh, make it executable with chmod +x review-changes.sh, and place it in your repository tools directory. The script below compares the current branch with main, limits the file list, and tells Codex exactly which output format to return.

#!/usr/bin/env bash
set -euo pipefail

BASE_BRANCH="${1:-main}"
REPORT_DIR=".ai-reports"
mkdir -p "$REPORT_DIR"

git fetch origin "$BASE_BRANCH" --quiet
FILES=$(git diff --name-only "origin/$BASE_BRANCH"...HEAD | head -n 80)

if [ -z "$FILES" ]; then
echo "No changed files found."
exit 0
fi

PROMPT="Review the current repository changes against origin/$BASE_BRANCH. Focus only on these files:
$FILES

Identify likely bugs, security issues, missing tests, and risky assumptions. Return sections named Summary, Findings, Tests, and Suggested Next Steps. For every finding, include filename, severity, and a concise fix. Do not modify files."

codex exec "$PROMPT" > "$REPORT_DIR/review-$(date +%Y%m%d-%H%M%S).txt"
echo "Review saved in $REPORT_DIR"

The important design choice is scope control. Passing a bounded list of changed files keeps the prompt relevant, reduces token cost, and makes the report easier to inspect. For large monorepos, narrow the file list further by excluding generated assets, lockfiles, build output, and vendor directories.

3. Make the Output Useful in CI

Run the script locally before adding it to CI. Review the saved report and tune the prompt using examples from your codebase. If Codex flags too many style issues, explicitly ask it to prioritize correctness, security, and production reliability. If it lacks context, include the pull request title, issue number, test command, or architectural constraints in the prompt.

• Use a read-only review prompt for pull requests.
• Use a separate branch and explicit approval for workflows that modify code.
• Set a maximum changed-file count to prevent unexpectedly large requests.
• Save reports as CI artifacts so reviewers can trace AI recommendations.
• Fail the pipeline only for clearly defined machine-checkable conditions, not subjective suggestions.

For scheduled maintenance, reuse the same pattern with a cron job or CI schedule. For example, replace the Git diff command with a dependency report, then ask Codex to identify outdated packages, breaking changes, and recommended upgrade order. The script remains the orchestrator; Codex provides analysis where rules alone are insufficient.

4. Control Cost Without Reducing Workflow Quality

Automation can generate many requests, so model routing matters. Use a fast, economical model for routine summaries and file classification, then reserve stronger models for security-sensitive reviews or complex refactoring plans. Because 59API supports Codex and standard OpenAI SDK integrations while offering inexpensive pay-as-you-go access, it is well suited to testing these workflows before expanding them across a team. Its referral rebate can also help reduce ongoing experimentation costs.

Start with one repository task, measure report usefulness, and refine the prompt every week. When the workflow consistently catches real issues or saves review time, move it into CI. If you want an affordable compatible endpoint for that first implementation, sign up for 59API and connect your existing Codex setup.

मूल्य · दस्तावेज़