# Show HN: Mob programming with AI agents, built on OpenSpec and OpenCode > Source: > Published: 2026-10-26 10:46:16+00:00 Click a story for its spec and activity or pick a goal Spec400123456789 Apple Pay and Google Pay buttons on the cart pageChrisSpec ready Welcome email with a 3-step setup checklistEmilySpec ready Rate limit the login endpointMarcusSpec ready Dark mode for the order history pageSarah Writing spec Build300123456789 ChrisSave addresses for returning customersChris Address picker in checkout EmilyInvite teammates from the empty dashboardEmily Invite form with email chips MarcusAudit log for admin actionsMarcus Audit log page for admins Review300123456789 Show shipping costs before the payment stepSarah Product tour for first-time sellersChris Export orders to CSV without breaking on accentsEmily Ship400123456789 Guest checkout without an accountMarcus Fix the typo in the footerSarah Sign in with GoogleChris Order confirmation email redesignEmily From intent to pull request on one board Agents that know the whole board Each card is a story. Agents read every open story before they plan. They skip work another story already covers. Write the story Describe the change the way you would brief a colleague and @mention whoever should weigh in. Got a bigger goal? An agent can break it into stories. You can also send them in from any tool through the project webhook. Read more ChrisSave addresses for returning customersChris Address picker in checkout EmilyInvite teammates from the empty dashboardEmily Invite form with email chips MarcusAudit log for admin actionsMarcus Audit log page for admins Apple Pay and Google Pay buttons on the cart pageChrisSpec ready Rate limit the login endpointMarcusSpec ready Dark mode for the order history pageSarah Writing spec Show shipping costs before the payment stepSarah Export orders to CSV without breaking on accentsEmily Guest checkout without an accountMarcus Fix the typo in the footerSarah Sign in with GoogleChris Order confirmation email redesignEmily Ship features by writing specs Griyd is spec-first. Your whole team writes each feature as a spec. Agents turn the spec into code and a pull request. An agent writes the spec Your agent reads the repository and every open story on the board. Then your agent writes the spec with OpenSpec. It covers what will change, the scenarios that prove it works and the tasks to get there. Not quite right? “Keep the old checkout behind a flag.” Your agent revises the spec. Agree on it, then Run Nothing runs until someone on the team presses Run. Fresh agents work through the tasks on the story's own branch, writing tests alongside the code. Watch the terminal and steer while they work. Read more Review with the full story Every task is checked off. Open the pull request with the whole history attached. Review the pull request Before the story reaches Review, every scenario in the spec is checked for code and tests. The branch is pushed, with the whole record on the story. Your automatic tests have to pass too. When a test fails, your agent fixes it. An agent-hour is an hour of sandbox runtime while an agent writes a spec, works through tasks or checks the result. Model usage is still billed directly by your provider on the keys you add. Your code stays yours Your keys Plain git Throwaway sandboxes Runs on Cloudflare Frequently asked questions What do we need to get started? A git repository with a personal access token, plus a model API key or a ChatGPT subscription. Add the repository as a project and write a story. The first spec arrives within minutes. Do we have to move our backlog into Griyd? No. Your tracker stays the source of truth. Send only the work you want agents to take through the project webhook. Each request becomes a story that gets its spec right away. Which languages and frameworks does it support? Any. Your agent reads the repository first so the changes follow your structure and conventions. Your CI checks the result the same way it checks your team's work. Can one goal span several repositories? Yes. Add each repository as a project and include them in the goal. An agent drafts stories for each repository in turn. Related stories see each other's specs so the pieces fit together. Do agents know about the other stories on the board? Yes. Agents read every open story before they write a spec. They leave out work that another story covers. When a story needs another one first, your agent names that story as a dependency. You can still run the story sooner. Can an agent merge or push to main? No. Agents can only push to their own story branch. Any other push is blocked before it reaches your git host. Merging always stays with a person. What if the agent gets it wrong? You review the spec before anything runs. While your agent works you can send a message or stop the run. If the pull request needs work, ask for changes on the story and run it again. Nothing merges until you merge it. Does the agent run our tests? Your tests run on your git host like they always do. Griyd waits for the result. When a test fails, your agent reads what went wrong and fixes it. Your agent tries up to three times. The story moves to Review only once all tests pass. If a test still fails, the story shows the error. Does Griyd keep our code or train on it? No. Each run clones into a sandbox that is deleted when the run ends. Griyd does not train models. Your code is processed by the model provider you pick under your own key. Who on the team can do what? Admins manage members and project settings. Members write stories, approve specs and run agents. Viewers can follow every story without changing anything. How much does a story cost? Model tokens are billed by your provider on the key of whoever presses Run. Sandbox time is free for five hours a month per person. After that your new runs wait for next month unless you upgrade to Pro at $0.75 per agent-hour. A single run stops after an hour.