Files
awesome-quant/docs/codex-setup.md
T
2026-06-29 17:07:52 -03:00

2.1 KiB

Codex Setup

This repo is ready for Codex through AGENTS.md and repo-scoped skills in .agents/skills.

What Codex Loads

  • AGENTS.md contains durable repository instructions.
  • .agents/skills/sprr reviews one PR.
  • .agents/skills/bprr reviews open PRs in bulk.
  • .agents/skills/update-pypi-dates refreshes tracked PyPI dates.

Restart Codex after adding or changing skills if they do not appear in skill selection.

GitHub MCP Requirement

PR review workflows require a GitHub MCP server. Configure it in your own Codex configuration, usually ~/.codex/config.toml. Do not commit personal access tokens or local auth files to this repository.

After configuring GitHub MCP, start Codex in this repository and run:

/mcp

Confirm that a GitHub MCP server is enabled and exposes tools that can:

  • list pull requests,
  • read PR details, files, labels, comments, and diffs,
  • read README.md from the default branch,
  • add comments and labels,
  • merge pull requests after explicit user approval.

If those tools are missing, $sprr and $bprr should report that PR operations are blocked instead of falling back to gh.

Local Validation

Install dependencies once:

uv sync --no-install-project

Validate README entries added relative to the base branch:

uv run python scripts/validate_readme.py --diff-from origin/main

Validate the full README:

uv run python scripts/validate_readme.py

Full validation reports legacy duplicate, URL, and format issues as warnings by default. PR-added entries treat those issues as errors.

Parser And Site Commands

parse.py needs a GitHub token because it enriches entries with GitHub metadata:

GITHUB_ACCESS_TOKEN=<token> uv run python parse.py

The site generator can run locally without a token when it can use existing CSV data or parse README.md directly:

uv run python site/generate.py

Claude Files

The existing .claude/ files and CLAUDE.md are kept for compatibility. Codex uses AGENTS.md and .agents/skills as the canonical workflow surfaces.