Skip to main content
All projects

Open source

Codex Skills Collection

Open-source multi-agent workflow system for Codex, built to make research, planning, implementation and review repeatable across delivery teams.

Repository
  • Codex
  • Agent workflows
  • Markdown
  • Shell

01Operating model

Process becomes part of the product.

The operating model turns an otherwise ad-hoc assistant setup into a delivery system: shared commands, routing-first skill loading and task-specific agent roles.

  • Research → Plan → Implement → Review
  • 66 reusable skills
  • 6 role-specific agents

02Context discipline

A system for choosing context, not hoarding it.

Skills are selected through a routing tree rather than dumped into context. That keeps the working set constrained to the current task and makes the setup portable between repositories.

  • AGENTS.md routing
  • Repository-aware discovery
  • Vault-grounded retrieval

03Adoption

Repeatability is the feature.

Bootstrap and sync scripts make the system practical to adopt: downstream projects receive the same commands, skills, agent templates and validation checks.

  • Dry-run bootstrap
  • Changed-only sync
  • Mechanical routing validation

AI Lab lens

Agent Workflows

Public workflow system for delivery teams using reusable Codex skills and command flows. It connects research, planning, implementation, review and handoff into one operating model.

Research question

How do reusable skills keep AI-assisted delivery consistent?
  • Researchtask-analysis + codebase-analysis skills gather context before any plan.
  • Knowledge Retrievalknowledge-retrieval pulls a bounded, justified context pack from a linked Obsidian vault before planning, never a full vault dump.
  • Planarchitecture-design + implementation-gap-analysis turn intent into ordered phases.
  • ImplementRouted through 7 domain branches, with Frontend splitting further by framework, to 66 route-validated skills.
  • Agent TieringSix role-specific agents, each assigned its own model, reasoning budget and sandbox permissions.
  • Reviewcode-review + security-hardening gate every backend change.
Browse all projects