Codex Context Kit

Codex Context Kit is an open-source skill pack for reducing wasted agent context. It helps Codex inspect large repositories, route work to the smallest relevant file set, and audit application LLM prompts with measured token usage.

The repo name is codex-context-kit: broad enough for more than token counting, specific enough to explain the job.

Skills

• repo-context-triage: Build a context budget before reading large files. Includes a small Git diff/stat script.
• focused-codebase-routing: Choose targeted files, tests, and commands in large or mixed-language repos.
• llm-runtime-token-audit: Audit app-side prompt builders, token telemetry, schemas, and payload compaction.

Install Locally

Clone the repo, then symlink the skills into your Codex skills directory:

mkdir -p "${CODEX_HOME:-$HOME/.codex}/skills"
ln -s "$PWD/skills/repo-context-triage" "${CODEX_HOME:-$HOME/.codex}/skills/repo-context-triage"
ln -s "$PWD/skills/focused-codebase-routing" "${CODEX_HOME:-$HOME/.codex}/skills/focused-codebase-routing"
ln -s "$PWD/skills/llm-runtime-token-audit" "${CODEX_HOME:-$HOME/.codex}/skills/llm-runtime-token-audit"

Restart Codex after installing or updating skills.

Validate

If you have the Codex skill creator validator installed locally:

scripts/validate-skills.sh

The skills are plain Markdown plus small helper scripts. No package manager setup is required.

Design Principles

• Measure first: inspect diffs, file sizes, dirty state, and token telemetry before reading broad context.
• Route early: decide the domain, entrypoints, tests, and generated paths before opening files.
• Prefer scripts for repeatable inspection and keep SKILL.md bodies short.
• Read test names and symbols before reading entire large files.
• Treat generated, binary, dependency, build, and cache paths as opt-in context.

License

MIT