A Claude Code wrapper that mirrors local work onto your Issues board — the same kanban, diffs, and cost badges as remote runs, driven from your own machine.
$ curl -fsSL https://agents.pinata.cloud/install | shOne line. No dependencies — just Node 20+.
$ curl -fsSL https://agents.pinata.cloud/install | shOpens your browser to sign in — no token copy-paste.
$ agents loginagents login on this machine…Maps the repo to a board, agent, and model.
$ cd your-project && agents initEvery task files a live issue on your board — diff & cost included.
$ agentsEvery command, plus end-to-end walkthroughs for the things you'll actually do.
Task-first walkthroughs. Each one is the shortest path to a real outcome — run the steps in order.
You just installed the CLI and want to point it at a project.
$ agents loginOpens your browser to authorize this machine — no token to copy.
$ cd your-project && agents initMaps the repo to a board, an agent, and a model.
$ agentsOpens the interactive prompt. Type a task and it files a live issue and starts working.
You have a concrete change in mind and want it filed, built, reviewed, and shipped.
$ agents start "add rate limiting to the templates route"Files a RUNNING issue and launches the agent on it.
$ agents pr 42 --draftPushes the branch and opens a GitHub PR (needs the gh CLI). Idempotent.
$ agents merge 42 --squashMerges the PR, flips the card to merged, and closes the issue as done.
You want a set of issues proposed from the actual state of your codebase.
$ agents file "split the oversized templates handler"Claude reads the branch, recent commits, and dirty tree, then proposes 1–5 issues. You confirm before any are filed.
$ agents ls --status backlogSee what landed.
$ agents work 47Pick one up and run the agent on it.
You need a fresh deployed agent with an API key wired into its environment.
$ agents agent create scout --engine superbuilderOr run `agents agent new` for an interactive prompt.
$ agents secret create OPENAI_API_KEY --type secretPrompts for the value privately if you omit --value.
$ agents secret attach OPENAI_API_KEY scout --as OPENAI_API_KEYApplies on the next restart.
$ agents agent restart scoutYou want an agent to run on your own Claude/OpenAI plan instead of a raw API key.
$ agents provider login claude --agent scoutRuns `claude setup-token` and saves the OAuth token, attaching it to scout.
$ agents auth --oauth openai --agent scoutAlternative: the browser OAuth flow for OpenAI/Codex.
$ agents provider lsConfirm what is connected and where it is attached.
You want to extend an agent with a published ClawHub skill.
$ agents clawhub list --sort popularBrowse the marketplace.
$ agents clawhub install <hub-skill-id>Pulls it into your library.
$ agents skill attach <name> scoutAttach one or more skills to an agent (applies on restart).
You want to talk to a deployed agent from a chat platform.
$ agents channel set scout slack --bot-token xoxb-… --app-token xapp-… --onTelegram, Discord, and WhatsApp work the same way.
$ agents channel ls scoutCheck each channel's status and whether tokens are set.
You want an agent to run a prompt on a cadence (OpenClaw engine).
$ agents task create scout --name digest --every 1h30m --prompt "summarize new issues"Or use --cron "0 9 * * *" / --at <ISO> for one-shots.
$ agents task run scout <jobId>Trigger it immediately to test.
$ agents task runs scout <jobId>Inspect the run history.
Your agent is serving something on a port and you want a public URL.
$ agents port start scout 3000 --cmd "npm run dev"Starts the dev server in the container and probes it.
$ agents domain add scout --port 3000 --subdomain scout-demoA subdomain is live immediately; a --custom domain needs the TXT + verify steps.
$ agents domain verify scout <domainId>Only for custom domains — checks the ownership TXT and SSL.
You want a restore point before a risky change — or to undo one.
$ agents snapshot create scoutSyncs the workspace to R2 as a new snapshot.
$ agents snapshot ls scoutList snapshots and copy the CID you want.
$ agents snapshot reset scout <cid>Rolls the workspace back (confirms first).
You want help from the platform assistant, or a quick reply from a deployed agent.
$ agents copilot "deploy a research agent and attach my OpenAI key"Tool-using assistant; it asks you to approve [y/N] before any mutating action.
$ agents agent scout "what are you working on?"Delivers a one-shot turn and relays the reply.
Run agents with no arguments to open a Claude Code–style prompt. Type a task to file and run it; follow-ups continue the same session and re-report the cumulative diff. Press esc to interrupt a turn. Inside the prompt, slash commands drive the board:
agents › add rate limiting to the templates route ↳ files a self-driven issue (RUNNING) and streams the work live agents #42 › also cover the x402 route ↳ same session, same issue — follow-ups continue the conversation agents #42 › /file split the oversized templates handler ↳ Claude drafts backlog issues; you pick which to file agents #42 › /new # wrap up → issue moves to REVIEW
/new | Wrap up the session → review. |
/work <id> | Pick up an existing issue as a session. |
/file [brief] | Draft backlog issues from context. |
/model [id] | Show or switch the model. |
/agent [name] | Show or switch the paired agent. |
/copilot [msg] | Switch the conversation to the platform copilot. |
/talk <name> [msg] | Switch the conversation to a deployed agent. |
/issue | Switch back to filing issues. |
/context | Show the working context. |
/mode [m] | Show or set the permission mode. |
/diff | This session's diff. |
/cost | This session's spend. |
/ls /open /comment /tag /done /reject | Board operations. |
/pr /merge /sync /board | PR + board operations. |
/clear | Clear the screen. |
/quit | Exit the REPL. |
Every command, grouped. Issue refs accept #numbers everywhere an id is expected. Most listing commands take --json for scripting, and a global --api-base URL overrides the host for one run.
agents login [--no-browser] [--token JWT] [--api-base URL]Sign in via the browser; --no-browser pastes a token, --token passes one directly. Verified on save.
agents logoutClear stored credentials.
agents whoamiShow the working context — identity, paired agent, board, model, repo.
agents initGuided setup: pick a board + agent, detect the repo, write .agents/config.json.
agents model [<id>]Show or switch the model (writes through to the board and the paired agent).
agents start "<task>" [--tag T]Create a self-driven issue + launch the agent; report on exit.
agents work <issueId>Pick up an existing issue and run the agent on it.
agents run "<task>"Headless: stream-json mode, full lifecycle unattended.
agents file ["<brief>"] [--tag T] [--yes]Draft 1–5 backlog issues from repo context; confirm before filing.
agents resumeRe-report the current diff for this repo's in-flight issue.
agents copilot ["<message>"]Chat with the platform copilot (tools + approve/deny). No arg → interactive.
agents agent <name|id> ["<message>"]Message a deployed agent and relay its reply. No message → interactive.
agents agent ls [--json]List your agents (name · id · engine · status).
agents agent create <name> [--engine e] [--template id] [--desc "…"]Create an agent. `agent new` is the interactive version.
agents agent get <name|id> [--json]Full detail — skills, secrets, snapshots, ports, domains, tasks, channels.
agents agent restart|status|rm <name|id>Restart · show status · delete (rm confirms).
agents agent exec <name|id> "<command>" [--cwd dir]Run a shell command inside the container.
agents agent logs <name|id>Print the tail of the container logs.
agents agent update <name|id> [check|apply] [--tag latest|0.3.0]Check / apply an in-container engine update.
agents agent versions <name|id> [--json]Current + available agent image versions.
agents agent engines [--json]List the engines this deployment has enabled.
agents secret ls [--json]List secrets with attached agents + mount aliases.
agents secret create <PATH> [--value V] [--type secret|variable]Create a secret (prompts for the value if omitted).
agents secret update <id|path> [--value V]Change a secret's value (keeps attachments).
agents secret attach <id|path> <agent> [--as ENV]Attach to an agent (applies on restart).
agents secret detach <id|path> <agent>Remove a secret from one agent.
agents secret rm <id|path>Delete a secret.
agents auth --oauth <openai|anthropic> [--agent A]Connect a provider subscription via OAuth (browser sign-in).
agents provider login claude [--agent A]Run `claude setup-token` and save the OAuth token (Claude subscription).
agents provider login codex [--agent A]Browser OAuth (PKCE) for the OpenAI/Codex subscription.
agents provider set <anthropic|openai|openrouter|venice> [--value KEY] [--agent A]Save a raw provider API key.
agents provider lsShow connected providers and attached agents.
agents skill ls [--json]List your skill library.
agents skill create --cid CID --name N [--desc "…"] [--env A,B]Register a skill in your library from a CID.
agents skill attach <cid|name> [<cid|name> …] <agent>Attach one or more skills to an agent (batched).
agents skill detach <cid|name> <agent>Detach a skill from one agent.
agents skill delete <cid|name> [--yes]Remove a skill from your whole library (≠ detach).
agents clawhub list [--category c] [--sort popular|newest|name] [--q text]Browse the ClawHub community marketplace.
agents clawhub get <slug>Show a hub skill's detail.
agents clawhub install <hub-skill-id>Install a hub skill into your library.
agents channel ls <agent> [--json]Show each channel's status (+ token-set state).
agents channel set <agent> <channel> --bot-token T [--app-token T] [--on|--off]Configure a channel (preserves enabled state unless --on/--off).
agents channel disable <agent> <channel>Turn a channel off but keep its tokens.
agents channel rm <agent> <channel>Remove a channel + clear its tokens.
agents task ls <agent>List scheduled tasks.
agents task create <agent> --name X (--every 1h30m|--cron "expr"|--at ISO) (--prompt "…"|--event "…")Create a scheduled task.
agents task update <agent> <jobId> [--name|--every|--cron|--prompt|--on|--off]Patch an existing task (only the fields you pass).
agents task run <agent> <jobId>Run a task now.
agents task runs <agent> <jobId> [--limit N]Show a task's run history.
agents task toggle <agent> <jobId> [--off]Toggle a task on/off.
agents task rm <agent> <jobId>Delete a task.
agents domain ls <agent> [--json]List subdomains / custom domains.
agents domain add <agent> --port N [--subdomain s] [--custom example.com] [--protected]Register a subdomain (immediate) or custom domain (needs TXT + verify).
agents domain update <agent> <domainId> [--port N] [--protected|--public]Change a domain's target port / auth.
agents domain challenge <agent> <example.com>Print the TXT record to add for a custom domain.
agents domain verify <agent> <domainId>Finalize a custom domain (TXT + SSL check).
agents domain rm <agent> <domainId>Remove a domain.
agents port ls <agent> [--json]List forwarded container ports.
agents port add <agent> <port>:<prefix> [--public]Forward a port at a path prefix (e.g. 3000:/app).
agents port start <agent> <port> [--path /x] [--cmd "<command>"]Start a dev server on a port and probe it.
agents port rm <agent> <port|prefix>Remove a forwarded port.
agents snapshot ls|status <agent> [--json]List workspace snapshots · R2 sync status.
agents snapshot create <agent>Sync the workspace to R2 (new snapshot).
agents snapshot reset <agent> <cid> [--yes]Roll the workspace back to a snapshot.
agents device ls <agent> [--json]List pending + paired device-pairing requests.
agents device approve <agent> <requestId>Approve a pairing request (or device approve-all).
agents config get|set|validate <agent> ['<json>' | --file path]Read · replace · validate the agent's engine config.
agents template ls [--category c] [--featured] [--json]Browse the template marketplace.
agents template get <slug> · template mineShow a template's detail · your own submissions.
agents template validate|submit <gitUrl> [--ref main] [--path dir]Validate / publish a template from a git repo.
agents feedback "<message>"Send a feature request / bug report to the Pinata team.
agents ls [--status S] [--board B] [--json]List issues — status, #, title, priority, PR, cost, tags.
agents open <issueId>Show one issue's detail.
agents comment <issueId> "<text>"Post a comment.
agents tag <issueId> "<a,b,c>"Replace the issue's tags (max 3).
agents done <issueId>Move an issue to done.
agents reject <issueId>Move an issue back to backlog.
agents pr <issueId> [--draft] [--title T] [--body B]Push the branch + open a GitHub PR. Idempotent.
agents merge <issueId> [--squash|--merge|--rebase]Merge the PR → mark merged + done.
agents sync [<issueId>]Refresh PR state from GitHub and print it.
agents upgradeUpdate to the latest published CLI (also repairs a broken install).
agents --versionPrint the installed version.
agents init resolves and persists the context every session runs under — shown in the REPL banner, agents whoami, and /context.
--model to every local run, so local and remote use the same model.~/.config/agents/credentials.jsonYour JWT, email, and host (mode 0600).<repo>/.agents/config.jsonBoard, repo, paired agent, model, permission mode.<repo>/.agents/state.jsonIn-flight issue + offline outbox (gitignored).--no-update, AGENTS_AUTO_UPDATE=0, or "autoUpdate": false in credentials.json. A failed self-update is non-fatal — your command still runs.