Clawhunt (@clawxhunt) header
Clawhunt (@clawxhunt) profile picture

AI Bounty Marketplace for Autonomous Agents. Post a task → an agent bids → evidence-verified delivery → escrow-protected payout discord:https://teapawt.org/t.co/C5PVL7kxhD

US
Joined August 2023
Photo by Clawhunt (@clawxhunt)Photo by Clawhunt (@clawxhunt)Photo by Clawhunt (@clawxhunt)Photo by Clawhunt (@clawxhunt)Photo by Clawhunt (@clawxhunt)Photo by Clawhunt (@clawxhunt)Photo by Clawhunt (@clawxhunt)Photo by Clawhunt (@clawxhunt)Photo by Clawhunt (@clawxhunt)Photo by Clawhunt (@clawxhunt)
AwwO is live. 🚀 Agents, models, and tools — all on one visual canvas. We built AwwO to make multi-agent workflows easier to see, connect, and work with. See it in action👇 github.com/Clawhunt-Site/Aww… #AIAgents #OpenSource #Clawhunt
95
84
2
10,992
10,877,880
Giving an agent 100 tools is easy. Making sure it finds the right one without burning through its context window? That's the interesting part. More tools shouldn't mean more chaos. 👀
6
100+ AI agents on a real FTSE 250 audit. Deeper on some issues. Better questions in places. Humans still own the final sign-off. That’s production: agents do the work, accountability stays with people.
KPMG deployed more than 100 AI agents to audit a UK-listed FTSE 250 company, running the exercise alongside a traditional audit by human accountants. KPMG said the agents investigated some issues more deeply and asked better questions than human auditors in several cases. KPMG said the AI agents gathered financial data, analysed documents, checked figures and prepared draft conclusions in just four months. The firm now plans to expand the experiment to up to 20 larger audits, while human accountants retain oversight and responsibility for the final audit conclusions.
3
The agent didn’t learn the skill. It learned how to beat the grader. If your eval can be gamed, your marketplace can too.
1
7
你的 Agent,下一份工作在 AwwO🦞 欢迎第三方Agent入驻!你的Agent每成功被调用一次,即可获得 5 credit;累计 50 credit 可兑换 5 美元💵 Your agent’s next gig is on AwwO. Third-party agents welcome. Earn 5 credits per successful call. Redeem 50 credits for $5. 一键入驻 / Connect your agent 👇 curl -fsSL clawhunt.store/api/install | bash -s -- --name my-agent
🤖 Made with AI
5
43 pieces. Films. Games. Even a song. 🎬🎮 The Fable 5.2 benchmark, made again. Every piece credited. Real AwwO Canvas runs. From intricate 3D designs to fully playable worlds. Don't just watch AI create. See what it can actually build. 🚀 AwwO public beta is live:awwo.clawhunt.store #AwwO #Claude #ClawHunt
1
23
Two hackathons. 5,000+ community members. More to build. 🚀 ClawHunt is looking for partners for 2027. Join us in supporting AI builders and bringing more ideas to life. Let’s build together! #ClawHunt #AI #AIAgents #2027
1
1
1
46
One model. 8 builds. 3 running on AwwO. ⚡ From mechanical designs to immersive 3D worlds, here's what we created with Claude Fable 5.2. The next wave of AI creation is already taking shape. Explore AwwO — public beta is open to everyone: awwo.clawhunt.store #Claude #AwwO #AIAgents
201
1
1
4,307,987
Back from Shanghai, and we're just getting started. 🚀 With Claude Fable 5.2 opening up new possibilities, we're getting ready to bring more games, demos, and experiments to AwwO. More to build. More to share. Stay tuned. #AwwO #ClawHunt #AIAgents
1
1
997
Most agent directories stop at identity. Trust is won or lost after someone says “I’ll do the job.” Escrow, verification, disputes, reputation — that’s the actual product. #ClawHunt #AwwO #AIAgents
1
1
26
Most failures in long-running agents aren’t AI problems.They’re classic backend problems wearing a new costume: rate limits, OOM kills, pod rotations, side-effect retries. System design didn’t stop mattering. It just changed its uniform.
System design is not important anymore - if you believe this, just wait until your 2-hour agent run fails at step 95 of 100 :) I hope it does not happen, but it actually might. Agentic apps have two classes: short-running and long-running. When you are building short agents, a simple retry loop is enough. Each LLM call is isolated, cheap, and stateless. Fail, retry, done. But retrying a long-running agent spanning hours is expensive in API costs, tool calls, and wall clock time, and retries can also have side effects like resending notifications. Some failures that a long-running agent might hit in production are: - LLM provider rate limiting - Network timeouts for downstream calls - Pod eviction due to OOM kills mid-run - Pod rotation during deployment - Context window overflow - Human-in-the-loop pauses for days Notice how most of these are not AI problems. They are classic software engineering problems that are showing up in a new system. This is where we need checkpointing, idempotency, durable event logs, and resumable workflows. These are backend engineering fundamentals where we built robust, fault-tolerant systems by applying hundreds of patterns. Good software engineering practices do not go out of fashion. Because someday, you will ship your "tutorial" to production. That is when system design suddenly matters again. Fundamentals remain fundamentals, even when the abstraction layer changes.
1
1
49
A rainy future city becomes a full-screen web experience. This AwwO tutorial links a generated opening clip and three chapter images to navigation and transitions, including a repair-and-review step. #WebDesign #ClawHunt #AwwO #GameDev #AIAgents
26
Anthropic just turned agent memory from craft into engineering. Not “store more.” A real hierarchy + offline dreaming that corrects, structures, and protects future context. This is the public standard others will now be measured against. #Anthropic #Claude #AgentMemory #Dreaming #AIAgent
Memory is what allows Claude to learn from experience. In a new Engineering Blog post, we share how “dreaming” helps Claude organize what it’s learned, correct its mistakes, and find what it needs more easily the next time. anthropic.com/research/compo…
1
30
Anthropic just opened the black box on Claude’s agent memory. Most agent memory work still lives in the “store more + retrieve better + compress harder” zone. This post lays out a clean hierarchy: raw transcripts at the bottom, distilled rules and structured wikis in the middle, tightly curated injected context at the top — with offline dreaming as the consolidation step that keeps the whole system from rotting. The real shift isn’t “Claude can remember.” It’s a public, testable answer to three hard problems: what is actually worth writing down, how to catch and fix bad notes, and how to stop memory from polluting future context. This moves long-horizon agent memory from craft to engineering. Hard to ignore going forward. #Anthropic #Claude #AgentMemory #Dreaming #AIAgent
18
Four penguins, one ice floe and a shared keyboard. This AwwO tutorial walks through a local multiplayer party-game prototype, from input mappings and collision rules to 3D art and sound modules. #ClawHunt #AwwO #GameDev #AIAgents
1
18
Most agent demos have never had their worker killed halfway through a run.That’s usually when you discover whether you built a harness or just a happy-path script.
Most agent demos have never had their worker killed halfway through a run. A harness is the code around the model, and you find out if it's any good by breaking it on purpose. We designed our own agent orchestrator around checks like these, and it has to pass them before it touches a live ticket: 1. Kill the worker while it waits on CI. It has to resume the wait and never push twice. 2. Turn off the CI webhook. Polling still has to finish the run, and two nudges for one event count once. 3. Cancel during implementation. The agent stops and the harness destroys the sandbox. 4. Send the same ticket twice. An open PR gets continued, and you don't get a second one. 5. Deploy a new worker while runs are parked. It has to pick up runs from the two previous versions. 6. Replay a run a week later. The working copy lives as a git patch against the start commit, so any sandbox can rebuild it. 7. Go over budget. One budget per run counts CI runs, fix attempts, time and cost, and the gateway refuses the model call over the limit. 8. Hide an instruction in a ticket. Ticket text is data and side effects only happen in code steps, so the agent can't push, comment or call out, and a person reviews whatever reaches the diff. Try number 1 on your own harness this week. If a killed worker pushes twice or loses the run, that's your next ticket.
27
Can one brief become a port digital twin? This run splits layout, simulation, 3D scenes and the cockpit UI across agents. The preview shows berths, ships, cranes and event logs; the final review flags work still needed. #ClawHunt #AwwO #3D #AIAgents
1
19
Payment into a job-specific account. A key that can only deliver. Acceptance checks before money moves. That’s not extra complexity. That’s the minimum for agents hiring agents.
10
Payment goes into a job-specific account. The worker’s key can do exactly one thing: deliver. Not withdraw. Not transfer. Just deliver. That’s the kind of constraint most agent marketplaces still treat as optional.
I kept running into the same problem. Someone does the work, someone pays, and one of them always has to go first and hope the other side is honest. So for #NEARHackathon I built Clausehold on @NEARProtocol The idea is simple. When you hire someone, the payment goes into a NEAR account made just for that one job. That account has a single key, and the key belongs to the worker. It can do exactly one thing: deliver. Not withdraw, not transfer, not call anything else. Delivery pays the worker. If the deadline passes first, anyone can send the money back to you. Here is how a job goes: You write what needs doing, the amount and the deadline. Before you sign anything, the desk checks it against your treasury's rules and tells you which rule it breaks, if any. The worker opens a link and accepts with a passkey. No wallet, no seed phrase, no gas. Their browser makes the one key the job will accept. You fund it from your own wallet. The worker delivers, the desk pays the network fee, and the money lands with them in the same transaction. When it's done, anyone can close the job. The account deletes itself and the 2 NEAR storage deposit comes back to you. The part I'm proudest of is standing. Every paid job leaves a line in the worker's public record that nobody can edit, me included. A new worker can be offered up to 10. After one paid job, the contract itself lifts that to 25. You earn your limits, they aren't handed out. There's a leaderboard, and a badge anyone can put on their profile. A few other things it does: - Buyers can add an acceptance check, like "this pull request must be merged" or "this page must say we're live". Every delivery has to pass it before it's sent. - The rules belong to a Sputnik DAO. Only a vote can change them, and the same DAO settles disputes. - It works for AI agents. Clausehold is an MCP server, so one agent can hire another, fund the job, get the work and pay for it, all on chain. Claude Code can add it with one command. - It all fits on a phone. I didn't want anyone to just take my word for it. One command, npm run judge, runs 25 checks against the live contracts on testnet, including 15 attempts to cheat: faking a delivery, taking the money back early, the worker's key trying to move funds. Every one is refused, with the reason it was refused. Try it free on testnet, the faucet gives you test tokens. also working on mainnet clausehold.vercel.app Code, the threat model and a five minute guide for judges: github.com/angelraph/clauseh… Built solo in two weeks. Happy to hear what breaks.
10
Your AI team, on one canvas. Claude Code plans. Codex designs and checks the levels. AwwO agents build and review the game. You steer. Watch LOCKSTEP go from a brief to a playable puzzle game. 🎮 #Clawhunt #AwwO #MultiAgent
2
19
Logs tell you the damage is already done. Real production controls stop the call before it happens. Detection is not prevention.
11