Codex-native skills
Install the pack into $HOME/.agents/skills or repo-local .agents/skills and keep Codex close to real engineering, business, and office operations work.
Open-source operating kit for OpenAI Codex and ChatGPT
CodexKit turns prompt chaos into a repeatable work operating system.
A Codex-first starter repo with installable skills, templates, and workspace kits for engineering, project management, finance, legal, HR, operations, strategy, analytics, marketing, and CX.
81
Codex skills
9
Prompt playbooks
10
Automation recipes
10
Templates
6
Starter workspaces
Starter commands
START-HERE-WINDOWS.cmd
bash ./START-HERE.sh
node ./scripts/validate-pack.mjs
npm --prefix web run devSuggested rollout
- Install the skills into $HOME/.agents/skills or repo-local .agents/skills, then restart Codex if they do not appear.
- Choose one starter workspace that matches your function before you add more process.
- Use $skill or /skills when you want a specific workflow instead of relying on description matching.
- Adopt one high-reasoning skill and one low-reasoning automation skill before expanding further.
- Introduce automations only after output shape, owner, and skip logic are clear.
Two-speed operating model
Use high-reasoning skills for decisions and analysis, then pair them with low-reasoning automation skills for routine coordination work.
Operational guardrails
Automation recipes, department templates, starter workspaces, and MCP guidance focused on control, not novelty.
Codex operating model
CodexKit starts with how Codex actually gets used.
Instead of inventing fantasy agent hierarchies, the kit centers the real working surfaces teams depend on: discovery, analysis, writing, decision support, delivery, bounded delegation, and recurring background work.
Clarify mode
Frame the question, decision, or workflow before producing output so the deliverable serves the real business need.
Best for: Ambiguous requests, incomplete data, stakeholder alignment
Execute mode
Turn inputs into a concrete deliverable such as a plan, brief, analysis, template, operating pack, or code change with scoped verification.
Best for: Drafting, synthesis, implementation, analysis
Review mode
Audit a document, workflow, metric pack, deliverable, or code change with a findings-first risk model.
Best for: QA, approvals, red-team checks, release readiness
Decision mode
Reduce options, compare tradeoffs, and recommend a path leaders or operators can act on quickly.
Best for: Budget, vendor, policy, project, and strategy choices
Delegation mode
Break a larger problem into bounded Codex jobs or parallel workstreams with explicit ownership and integration points.
Best for: Parallelizable work that does not share a write scope
Automations
Schedule recurring Codex work with prompts, cadence, owners, and low-noise guardrails.
Best for: Recurring reporting, risk checks, documentation drift, routine analysis
What ships
Installable assets for knowledge work and delivery
Skills ship as plain folders with SKILL.md, optional agents/openai.yaml, and reusable guidance for engineering, project governance, executive communication, finance, legal, people ops, supply chain, strategy, analytics, marketing, CX, and low-reasoning office work.
Portable workflows for ChatGPT users
Playbooks, templates, and starter workspaces work even when local skill installation is not available, which keeps the repo useful across the wider OpenAI workflow surface.
Starter kits for real teams
Department workspaces give PM, finance, HR, legal, ops, and marketing teams a working spine instead of a blank folder and a vague prompt.
Explore next
Skills
Browse the installable CodexKit skill pack.
Templates
See department-ready outputs for routine work and operating reviews.
Workspaces
Start from a packaged workspace instead of inventing one from scratch.
Playbooks
See how structured prompts support analysis, review, delivery, and release work.
MCP
Adopt integrations deliberately instead of bolting on every server.