Plans, architecture & agents
Version 1.1.1 · September 24, 2026
Create focused Product Plans, practical System Architecture, supervised Agent Teams, and connected engineering handoffs. Each relevant artifact can link to the corresponding Codelit templates without pretending the generated draft has been saved online.
Source published at codelit-io/codelit-plugin, verified public on September 24, 2026. This contains the plugin skills, public references, templates, tests, and branding—not Codelit's production application.
Publishing this source repository does not submit, approve, or publish the plugin in an OpenAI directory. The OpenAI directory submission remains a draft. The 27 synthetic behavioral cases passed; see the measured results and limitations below. See VALIDATION.md for the exact scope of completed local checks.
| Skill | Purpose |
|---|---|
| codelit-guide | Explain Codelit and route the request. |
| codelit-product-plan | Scope a first release, requirements, and acceptance criteria. |
| codelit-architecture | Explain or design components, data flows, boundaries, and recovery. |
| codelit-agent-team | Define a minimal supervised roster, permissions, handoffs, and evaluations. |
| codelit-plan-and-ship | Connect reviewed decisions into implementation tasks and tests. |
These are instruction workflows, not five independently running agents.
The displayed name is Codelit. The stable internal package name is codelit-copilot. The original 600 × 600 PNG supplied by the publisher is used unchanged for the directory logo and composer icon. Both manifest paths resolve to assets/codelit-logo.png.
Product Plans link to https://codelit.io/specs; Architecture to https://codelit.io/templates; Agent Teams to https://codelit.io/agent-templates; and connected handoffs to https://codelit.io/plan-and-ship. The rules live in template-links.md.
The plugin supplies the useful artifact first. It does not hide the result behind a signup requirement, add private content to URLs, invent template slugs, or represent an unsaved draft as a saved artifact. Each independently exported asset carries its appropriate link unless the user asks for no links. An exact public template is preferred only when its URL is verified. Simple explanations and unrelated replies need no navigation footer.
No Codelit account, repository, external app, MCP server, model key, analytics collector, schedule, installation hook, or production runtime is connected by this package. It does not save projects, run independent agents, deploy, or send messages merely because it is installed. Available host tools retain their own authorization rules. If the user opens a Codelit link, normal website requests occur; the plugin does not embed the user's draft in that URL.
Public documentation does not prove Codelit's private implementation. Proposed templates are not official import schemas. External actions need observed confirmation before success is claimed. A timeout after a write is an unknown outcome, not permission to repeat it blindly.
plugin.json is the portable manifest. .codex-plugin/plugin.json is the compatibility overlay for the same identity. There is exactly one plugin root.
Run python3 scripts/validate_bundle.py from the extracted directory. It uses Python's standard library and performs local, static checks without network calls. See VALIDATION.md for the recorded scope. It does not run a language model or certify OpenAI approval.
The behavioral cases and reviewer cases were executed against the installed plugin. See measured results for all 27 outcomes and the bounded-test limitations. Public-review preparation is in the submission checklist, listing copy, reviewer cases, and release notes.
Create a Product Plan for a client onboarding app. Keep the first release small, mark assumptions, add acceptance criteria, and link to relevant Codelit starters.
https://developers.openai.com/plugins/build/plugins https://developers.openai.com/plugins/deploy/submission https://developers.openai.com/plugins/deploy/submission-errors
Use Git, GitHub CLI (gh), and Python 3.9 or newer. Authenticate to GitHub using gh auth login and configure your own Git commit name and email. The authenticated account needs permission to create a public repository in codelit-io.
bash scripts/publish_github.sh --dry-run
bash scripts/publish_github.shThe helper targets only codelit-io/codelit-plugin and creates it with public visibility. It refuses to overwrite an existing repository or reuse an existing Git worktree, runs local validation, creates a new initial commit, publishes through gh repo create, and compares the remote commit with the local commit. It does not configure a live Codelit integration or submit anything to an OpenAI directory. A failed or timed-out create/push must be inspected before any retry.
No open-source license has been selected or added to this package. A public-repository request is not treated as a choice of MIT, Apache, or another license. The publisher should select the intended code/content license and separate brand-use terms before inviting third-party reuse.
See the connection guide and real ChatGPT Developer Mode demo for setup, example prompts, troubleshooting, and support. The public MCP server is available now; the ChatGPT directory plugin is submitted for review and is not yet publicly listed.
A separate read-only MCP service is available at https://codelit-public-mcp.vercel.app/mcp. It lists public planning workflows and retrieves bundled planning templates. The published 1.1.1 ZIP remains the tested skills-only package; this service is a separately deployed companion for the OpenAI submission. It does not access Codelit accounts or execute plans.