Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.
JavaScript 3d-generationai-agentsanimationavatardigital-humangenerative-aiglbgltfimage-to-3dllmmcpmcp-servermodel-context-protocolsolanatext-to-3dthreejsweb-componentswebglwebxrx402
docs(changelog): tell holders the AR page was printing commands the glTF tool rejects
nirholase1bcd08d37a2
diff --git a/CHANGELOG.md b/CHANGELOG.mdindex 82a53451d..d6d6453c0 100644--- a/CHANGELOG.md+++ b/CHANGELOG.md@@ -9,12 +9,15 @@ Public history for [three.ws](https://three.ws), newest first. New pages come fr- **Agent Interaction Lab** (`/demos/agents`): Fourteen prototypes exploring how a 3D agent physically inhabits a web page: real gravity, page elements as the playground, and the agent reacting to scroll, clicks, and the cursor.- **Docs · Draft Agent Mint** (`/docs/draft-agent-mint`): Every finished reconstruction is stored durably and minted as a draft on-chain agent identity: the Solana leg, the ERC-8004 leg, the flags that arm each one, and how to prove the whole path on devnet.- **Docs · Likeness evaluation** (`/docs/likeness-eval`): How three.ws measures whether a generated avatar actually looks like the person in the photos: three head-framed renders, an open-source face-recognition embedding, and a 1-5 score anchored to the model's own same-person threshold rather than to anyone's opinion.+- **Docs · pump.fun first-claims channel** (`/docs/pump-claims-channel`): A live Telegram feed of first-time pump.fun creator fee claims: how the delivery lane scans, dedupes by transaction signature, and drains a backlog in order, plus the configuration it needs and why the direct on-chain fallback cannot serve the feed.- **Docs · Recurring payments** (`/docs/recurring-payments`): Sign one permission, then a fixed USDC amount moves every day or week, or gets swapped into a token on a schedule. The lifecycle, the API, what every kind of failed charge does to a schedule, and the operator config settlement needs.+- **Docs · Site load performance** (`/docs/site-performance`): The load-cost rules behind every page that is not /play, each with the Lighthouse measurement that produced it: catalogue grids skip the rows nobody can see, analytics waits for the first idle moment, the CDN bundle ships minified, below-the-fold sections load below the fold, and thumbnails decode off the critical path.- **Docs · The Agent Index** (`/docs/agent-index`): The platform's own cross-chain index of agent registrations, token launches, reputation attestations, transfers and delegations, Solana first: one public endpoint, absolute on-chain timestamps, and published index freshness.- **3D collectible lookups reach more chains, and scene minting stops naming a missing secret**: The endpoint behind the in-chat 3D NFT viewer kept its own hand-copied list of which EVM networks it could look a collectible up on. The copy had drifted from the list the rest of the platform uses: two test networks were missing outright, so a collectible on either was refused as an unsupported network before anything was tried, even though the viewer can read those tokens straight off the chain without an indexer at all. The endpoint now derives that list from one shared source, so the two cannot drift apart again, and it admits any network the platform can actually reach. It also honors per-network indexer keys the way the rest of the stack does, instead of reading only the single shared key, and when a lookup cannot be indexed it now says which of the two reasons applies. Separately, the scene-minting endpoint answered anonymous callers with the name of the storage secret the deployment was missing, and did so before spending a rate-limit token; it now requires a signed-in minter first, so that message reaches only the people it is meant for. `[fix, security]`- **A batched order on the OKX Agent Identity Studio can no longer be paid at the price of one identity**: The Agent Identity Studio accepts several calls in a single request, and its price is the sum of every paid call in that request. The OKX X Layer payment option, the one an OKX buyer's tooling selects first, was quoting a flat one-identity price no matter how many identities the request actually ordered, so a multi-identity order could be paid at the price of one. Every payment option on that endpoint now quotes the same total, and a request with nothing chargeable in it still quotes the normal list price. Two smaller fixes ride along: a paid response now lets browser-based buyers read the settlement receipt header it has always sent, and the free catalog and health endpoints answer errors in the same shape as the paid endpoints beside them, with an Allow header naming the methods they take. (`/docs/okx-marketplace`) `[security, fix]`- **A bounty that does not exist now says so, instead of blaming pump.fun**: The bounty board at /bounties mirrors the public pump.fun GO board. When you opened a bounty link that had been deleted or mistyped, the page reported it the same way it reports pump.fun being down, so there was no way to tell a dead link from an outage. A bounty that genuinely is not there now answers as not found, and a real upstream problem keeps its own separate answer. A second, quieter case got fixed with it: when the submissions list for a bounty failed to load, the page showed the bounty with zero submissions and said nothing, which looks identical to a bounty nobody has entered. That failure is now recorded, so a pump.fun outage shows up as an outage rather than as a board that suddenly looks unpopular. (`/bounties`) `[fix]`- **A database blip during a scheduled job now reads as a database blip, not a healthy run**: Two of the scheduled jobs behind the trading fleet could fail without anything looking wrong. The job that reallocates budget across strategies talks to the database over a socket, and when that socket never opened the failure arrived as an object carrying no message at all. The job caught it, answered with an empty error string, and the scheduler recorded a success, so the health record stayed green while the loop did nothing. Failures of that kind are now recognised as what they are, a database that could not be reached, so the run is marked unhealthy and retried on the next tick instead of being logged as fine. Every page and endpoint that reads from the database inherits the same correction: this class of outage now returns a retryable service-unavailable response rather than an internal error. The fleet watchdog also gained a seventh check, over the calibration pass that measures whether a high-conviction coin actually wins as often as its score claims, so that loop can no longer go quiet unnoticed. `[fix, infra]`+- **A dedicated channel for first-time pump.fun creator fee claims**: The first time a coin creator pulls their accrued fees out of the vault is an irreversible on-chain signal that they are actually engaged with what they launched, and it was only visible by polling an API. There is now a delivery lane that posts each of those first claims to its own Telegram channel as it lands, one message per claim, with the creator, the amount in SOL, and links to the transaction and the coin. It keeps its own state so switching it on never dumps a backlog into the channel, it drains a backlog in order rather than skipping ahead if it falls behind, and it recognises claims by transaction rather than by timestamp, so two claims landing in the same second both get posted instead of one being silently dropped. The feed stays quiet until a claims indexer is configured: the direct on-chain fallback samples under a second of activity on a program this busy, which is not enough to tell a first claim from a repeat one. `[feature, infra]`- **A failed database migration can no longer freeze the schema in silence**: Every deploy carries database migrations that must run before the new code does. When one of them failed, the runner printed a single line with the database's own error and stopped, naming neither the file that failed nor the migrations queued behind it. Those stayed unapplied and unmentioned, so the site ended up serving code that reads columns the database never got, and the first visible symptom was an unrelated-looking error on a live page. The runner now names the failing file, how many ran before it, and every migration still waiting, and the deploy step that catches a database running behind the code is now part of the documented deploy path instead of something you had to remember to run. `[infra, fix]`- **A paid-call share link now previews on Slack, Telegram and LinkedIn**: Every settled x402 call has a permalink whose social preview image is drawn on demand. Slack, Telegram and LinkedIn ask an image URL for its type and size before they fetch it, and this one answered "method not allowed" to that probe, so the card silently never appeared on those platforms. It answers them now. Three more fixes to the same card: a receipt is captioned with the network it actually settled on rather than always reading mainnet; a link shared in the seconds before its receipt finishes saving, or during a cache outage, is retried a minute later instead of being pinned to the generic card for an hour; and a damaged record shows a plain "unknown" rather than a "NaNd ago" timestamp or an invented amount. (`/pay`) `[fix]`- **Add funds now opens Coinbase on the buy step, not on a transfer you cannot make**: The Add funds window hands you a Coinbase checkout with your three.ws wallet already filled in as the destination. That checkout was not told which step to open on, so Coinbase could land you on "transfer from your Coinbase balance", which is no help at all if you reached Add funds precisely because you hold nothing yet. It now always opens on the buy step for the asset you picked (USDC or SOL, both settling on Solana). The suggested dollar amount is also rounded to real dollars and cents before it reaches Coinbase, so an amount worked out from a shortfall can no longer show up as something like $25.777777 on the checkout screen or in the window itself. `[fix, improvement]`@@ -42,6 +45,7 @@ Public history for [three.ws](https://three.ws), newest first. New pages come fr- **Every avatar you scan now arrives with an on-chain identity of its own, minted for you in the background**: When a scan or a prompt finishes building your avatar, the finished model is written to permanent storage and the agent behind it is given a real on-chain identity at the same moment. It is a Metaplex Core asset on Solana, minted into the platform's own agent collection with the royalty terms and registry enrolment every other platform mint carries, and it is owned by the agent's own wallet rather than by us. There is no wallet prompt, no fee, and no step to remember: by the time the avatar shows up in your library the identity already exists and anyone can look it up. It is created as a draft, which means it says the agent exists and this is its avatar, not that it is live and taking work. Going live stays a deliberate choice you make later. If the mint cannot run for any reason your avatar is completely unaffected, because it is already delivered and saved before the mint is ever attempted. Agents that want an Ethereum-side identity as well can have one registered under the ERC-8004 standard, and the whole path is now documented, including the single setting an operator flips to arm it for the main Solana network and the script that proves the whole chain end to end on the test network. (`/create/selfie`) `[feature, docs]`- **Every Oracle card now says why that coin scored, and what a score like it has actually been worth**: The live conviction board had a copy problem that made a working engine look like a broken one. Each card carried a one-line read of the launch, and that line was assembled from whichever of the four pillars scored highest. Buy-side momentum is the highest pillar on about nine in ten of the coins at the top of the feed, so nine in ten cards printed the same sentence, word for word, down the whole page. Two hundred cards, one take, nothing learned. Meanwhile the engine had already written a specific, quantified reason for every single coin, and it was only visible if you opened the coin. Those reasons now ship with the feed, so each card quotes its own evidence and the historical lift behind it: 40+ early buyers at 6.8x the base rate, top holder at 5-15%, more sellers than buyers. Each reason is phrased in the units a trader thinks in rather than the model's internal labels. Every card also carries a second, quieter number: how often a coin scored into that band went on to graduate or run 2x or more without ever rugging, measured across the 61,916 scored coins the market has already resolved. That is deliberately a stricter question than the score answers, because the score ranks the odds of a run and says nothing about surviving it, and a holder is entitled to both numbers rather than being left to read a conviction of 99 as a 99 percent chance of anything. Two badges were cleaned up in the same pass: the prime pill that only ever repeated the tier printed beside it is gone, and the momentum badge, which was firing on 93 percent of the live feed, now appears only when buy pressure alone would carry a launch to prime. (`/oracle`) `[improvement, fix]`- **Keyboard users stop tabbing past the same Skip to content link twice, and the homepage hint is readable again**: Two accessibility defects on high-traffic pages are fixed. The shared navigation injects a Skip to content link as the first thing you reach with the Tab key, so keyboard and screen-reader users can jump straight to the page body instead of walking the whole header. It only checked for its own copy of that link before injecting, but roughly three dozen pages write the anchor into their own markup under a different class, so those pages ended up with two identical skip links and you had to tab past the same control twice to get anywhere. The check now recognises every variant, so there is exactly one skip link per page. On the homepage, the hint under the interactive dragon that reads move cursor, hold to breathe fire was drawn in white at 18 percent opacity over a near-black panel, about 1.6 to 1 against the background where the accessibility standard asks for at least 4.5 to 1; anyone in bright light or with reduced contrast sensitivity simply could not see the instruction for the thing it was explaining. It now sits at 5.7 to 1 and stays legible without changing the look of the scene. The Genesis and Walk pages also had a skip link with no styling rule attached, which left it as stray visible text above the header; both now use the shared style and hide until focused like every other page. `[fix]`+- **Links inside the tutorials now go where they say, and the AWS page names the right region**: Every tutorial on the site is written as a markdown file that lives beside the other documentation, and the links inside it were written against that location. The page that renders them was reading those links as if they pointed at site addresses instead, so a tutorial that sent you to a related guide, or cited the exact source file a step describes, sent you to an address that does not exist. Links that reach across to another document, that point at a specific section of one, or that cite a file in the source tree are now each resolved to the right destination, with source files opening on GitHub. The scene tutorial's own list of related tutorials pointed at a secondary address that rendered them under a generic title, and now uses the real one. Separately, the AWS page claimed the platform's production deployment runs in AWS us-east-1. It does not: production runs on Google Cloud Run, and us-east-1 is where the AWS Marketplace billing and entitlement APIs are called. The page now says that instead. `[docs, fix]`- **Oracle pages now say the data is briefly unavailable instead of showing you a blank market**: When the database had a short connectivity blip, the Oracle read endpoints answered with an empty result and a 200 OK, which reads as fact rather than as an outage. The live conviction feed showed a market with nothing in it, the hot-sectors panel showed no sectors, an agent's public track record showed "never traded" (the very evidence copy-traders judge a leader on), and the follow endpoint told a subscriber they were not following an agent they were in fact following. Worst of all, unfollowing an agent answered "done" while the subscription survived, so the alerts kept arriving with no way to stop them. Some of those answers were then cached for up to five minutes, outliving the blip that caused them. All of these now return a proper "temporarily unavailable, retry shortly" response that nothing caches, so a page can tell you it cannot reach the data instead of quietly showing you the wrong answer. `[fix, improvement]`- **Paid 3D downloads work from a browser again, and the marketplace report stops ignoring every Solana price**: Buying a 3D asset or an animation clip from a web page was impossible: those two endpoints answered the browser's permission check with a payment demand and none of the cross-origin headers a wallet needs, so the page could never even read what it was being asked to pay. Both now answer that check the same way every other paid endpoint does, on the challenge, on the errors, and on a free listing. The paid marketplace report had a second, quieter problem: it recognised prices set in USDC on Base but not the identical prices set in USDC on Solana, so every Solana-priced agent was counted as unpriced and left out of the average, lowest and highest figures buyers were paying for. Solana prices now count. The auth health check no longer charges for a call it refuses, its request reader now has the same size limit as its neighbours, and the avatar optimization batch can finally be repriced at deploy time like every other endpoint. (`/docs/x402-endpoints`) `[fix]`- **Paying to resolve a name that does not exist no longer costs anything**: The paid name-resolution call on /api/x402/pay-by-name sells one thing: the on-chain address behind a @username, a .sol name, or a raw wallet. When the name resolved to nothing it still took the $0.001 toll and handed back an empty answer. It now returns a plain 404 and leaves the payment unspent, so a buyer only ever pays for an address they actually received. Two smaller corrections shipped alongside: the pump launch feed's headline fields (newest mint, name and symbol) kept describing the newest launch when the list is ranked by liquidity instead of by age, rather than whichever coin happened to sit at the top; and internal and subscription callers can once again reach the Permit2 gas-sponsoring demo on deployments that have no card-facilitator credentials, since those callers never sign a payment in the first place. (`/docs/x402-endpoints`) `[fix]`@@ -57,10 +61,13 @@ Public history for [three.ws](https://three.ws), newest first. New pages come fr- **The activity floor stops reporting a flat week when nothing has closed yet**: The header on the live agent activity page counted a week of wins, losses and profit from the moment each position was opened, not the moment it resolved. A conviction position routinely takes days or weeks to settle, so by the time a trade had a result it had already aged out of the window that was supposed to count it. The tiles were therefore pinned at zero wins, zero losses and a confident PnL of plus zero forever, which reads as a floor that traded all week and broke exactly even. Entries are now counted on when the agent acted and results on when they settled, and the endpoint also reports how many closed positions carry a realized figure, so a genuinely flat week is told apart from a week with nothing closed yet: the tile says not available instead of inventing a zero. The win rate is graded against the market's own base rate rather than a hardcoded fifty percent, so a hit rate several times better than a random launch stops being painted red. Every tile now explains what it measures on hover, the stats row no longer re-reads all nine numbers to a screen reader on every incoming trade, and a stats outage shows the real values it has instead of shimmering as though it were still loading. Two lists on the page, the movers and winners rails and the insight strip, had their two-line rows collapsing onto a single run-on line; both stack properly again and truncate long names instead of overflowing. (`/activity`) `[fix, improvement]`- **The agent directory describes itself correctly in all 84 languages again**: The English description Google and every social preview show for the agent directory was updated a while back to say the directory spans Solana and every supported EVM chain. The 84 translated versions were never updated with it, so a reader in Spanish, Japanese, Arabic or any other language got the older, vaguer line that named no chain at all. All 84 now carry the current wording, with Solana and three.ws left in their real spelling rather than bent to fit each language's grammar. Retired copy was cleared out at the same time: eight strings that no page shows anymore had been sitting in every one of the 84 catalogs, which is 672 dead entries the site was shipping to readers for no reason. (`/agents`) `[fix, docs]`- **The agent interaction lab is reachable again from the demos index**: The demos index has always carried a card for the agent interaction lab, the set of small pages where an avatar high-fives you, climbs the page title, bounces on a trampoline, or falls asleep when you leave it alone. Clicking that card gave you a 404. The lab hub lives in a folder rather than as a single page, and the rule that turns a demo name into a page file was sending the request to a file that never existed, so the one link into the lab was the only link on the index that could not be followed. The sub-demos were fine the whole time; only the door was missing. It is fixed, and the demo route map now records the whole lab, including the lost-avatar 404 page that has its own card on the index but was never listed. (`/demos/`) `[fix, docs]`+- **The AR docs stopped handing readers three commands the tool rejects or quietly ruins**: Eight pages were read line by line against the code and the live site: the agent abilities dossier, news, the changelog, AR viewing, BNB payments, Agent Sniper, earned autonomy, and the agent personality tutorial. Every command printed on them was actually run. The AR page was wrong three times, and each one cost a reader real time. Its model optimization step used a flag the glTF tool does not accept, so the command simply failed. The line beside it promised a lossless result, when that tool compresses and simplifies by default: the file it produced could not be opened by the iPhone and Android AR viewers the page is about, and nothing said so. Its troubleshooting step for exactly that failure repeated the same rejected flag, so the fix for the problem hit the problem again. All three now carry flags the tool accepts, verified by running them on a real model and reading back what came out, with a note about what the defaults do. The AR page also linked to a placeholder address rather than a page, and now points at a real one. On the earned autonomy page, the list of settings a trading arm unlocks once it has proven itself was one short, missing the size of the position it keeps riding after taking its stake back. Everything else held up, including all 4,854 links across the eight pages. (`/docs/ar`) `[docs, fix]`- **The Arena now runs a daily competition on its own, instead of waiting for someone to host one**: The Arena is where agents compete head to head on real pump.fun profit and loss, but it could only ever show a competition a person had created by hand, and nobody had. The page had sat on a single leftover test bracket with no entrants since June, while the homepage sent people straight to it. It now runs itself: a keeper opens today's Daily Arena and queues tomorrow's, enters every agent that is actually trading, and closes out any competition whose clock has run out, freezing the final ranking and publishing it on-chain. That last part also covers competitions people host themselves, which previously waited on their creator forever. Two smaller cleanups came with it: a finished bracket nobody entered is no longer listed, since it has no result to show, and closing an empty bracket no longer publishes an on-chain record of nothing. (`/arena`) `[feature, improvement]`- **The Arena opens on the live board instead of a list, and a shared bracket now previews the standings**: Landing on the Arena used to mean reading a grid of cards before you could tell whether anything was happening. Whatever competition is running now is promoted to the top of the page as a full board: the podium with each agent P&L for the window, the countdown to the bell, and how many agents and trades are in it, all before a click. If nothing is live, the next scheduled bracket takes the slot. Sharing improved too. A bracket link pointed at a part of the address that link previews cannot read, so posting a live competition looked exactly like posting the homepage. Every bracket now has its own shareable address that unfurls a picture of the actual standings, drawn from the same numbers the page shows and the same ones published on-chain. A losing board is drawn exactly like a winning one, in red. (`/arena`) `[feature, improvement]`- **The CA to x402 tool stops handing out a command that cannot run**: Paste a contract address into the CA to x402 tool and it generates the copy-paste code for calling the paid endpoint it just built for you. The cURL tab's second step told you to pay by running npx x402-fetch against the endpoint. That never worked: the published x402-fetch package is a library, not a command line tool, so anyone who copied it got an npm error about not being able to determine an executable to run, on their first attempt to pay us. The tab now shows the real two-step HTTP flow instead, taking the 402 and then replaying the same call with the payment header, and says outright that the Node and Agent tabs are what sign that header for you. Both commands were run end to end against the live endpoint before shipping. The docs page carries the same correction, and the vault and master wallet docs pages got proper search descriptions in place of ones that quoted raw filenames at readers. (`/ca2x402`) `[fix, docs]`+- **The choreography docs no longer teach an example that quietly does nothing**: A routine step can pin one specific animation clip instead of letting the gesture slot pick one, and the choreography reference showed how, with a shareable link that pinned the dance slot to the rumba clip. That example never pinned anything. Rumba is already the platform default for the dance slot, and the runtime deliberately drops a clip that matches the slot's own default, so that a shared routine still adapts to an agent that has remapped its gestures. Anyone copying the example got a step that resolved normally instead of the pinned motion they asked for, with nothing to indicate why. The example now pins a clip that is not the default, and the field reference states the drop rule outright. The selfie-to-avatar page also had a summary line quoting an older, lower head-coverage number than the measurement reported further down the same page, and the avatar-thumbnails page printed a sample coverage readout from a time when three quarters of avatars still had no preview image. Both now show figures taken from the running system. (`/docs/choreography`) `[docs, fix]`- **The commit and changelog feeds can no longer repeat a post they already sent**: The jobs that deliver commits and changelog entries to the holders' Telegram channel take a lock so two overlapping runs cannot announce the same thing twice, and that lock lasts four minutes. But the calls they make to GitHub and Telegram had no time limit of their own, so a single stalled call could outlive the lock it was holding, let the next run pick the lock up, and post the same commits again. Every outbound call in those jobs is now bounded well inside the lock's lifetime, so the no-duplicates guarantee holds even when an upstream service hangs. The same missing limit is fixed across four other scheduled jobs: the wardrobe catalog audit, the garment job sweep, and the weekly quality benchmark, whose own run budget was previously advisory because it was only ever checked between requests and never during one. A stalled status check no longer fails a generation that is still healthy, and a slow download is no longer reported as a corrupted garment. (`/changelog`) `[fix, infra]`+- **The commit feed went quiet for thirteen hours, and the reason it could not restart itself is fixed**: Every commit that lands on the platform is posted to the holders' Telegram channel as it happens. That feed stopped at 06:45 UTC and stayed silent for the rest of the day, and the cause was in how it found its place again after each run. It looked back over only the thirty most recent commits for the last one it had announced. On a day when work lands in bursts of twenty or thirty at a time, that marker regularly fell out of view, and when it did, the feed quietly skipped to the newest commit and posted nothing at all. Every commit in the gap was passed over with no record that anything had been missed, and because each run repeated the same skip, the feed had no way back on its own. It now pages back as far as five hundred commits to find its place, and when it is behind it works forward from the oldest commit it still owes rather than jumping to the front, so a backlog is delivered in order over the following runs instead of being discarded. Anything more than three days old is still left alone, so catching up can never flood the channel with history. The 117 commits missed during today's outage are not recoverable and will not appear. `[fix]`- **The contracts that hold agent earnings, reputation stakes and skill licenses are now audit-ready**: Before an outside firm can review the contracts behind agent royalties, reputation staking and wallet delegation, someone has to write down what those contracts promise and then prove it holds. That pack now exists and is public. Every value-bearing contract has numbered invariants, each backed by a test proving the property holds and a second test proving that an attacker trying to break it is rejected, alongside a threat model naming what is worth stealing, who could try, and what stops them. The Solidity contracts sit at 100% test coverage across 242 tests. The two Solana programs, which had no automated tests at all, now have 32 running against their real compiled on-chain bytecode rather than a stand-in, with the real token programs loaded. Two standard analyzers were run across everything, and every finding is either fixed or answered in writing. One of those fixes closes a real gap: a cross-chain permission request that was declined rather than failed outright would have credited a seller and charged a buyer for access that was never going to be granted. `[security, infra, docs]`- **The crypto knowledge base the agents read from only holds stories we could actually read**: The news reader recorded every URL it was handed into the durable corpus the 3D agents ground their crypto answers on, including ones no extraction attempt could read. That left headline-only rows with no article behind them, and because the reader is a public endpoint whose caller supplies the link, the headline, and the source name, a single request could plant a permanent entry with any headline someone chose. The reader now refuses anything that is not an ordinary web link, records only stories that yielded real article text, and the knowledge API filters the empty rows already on file out of every answer. (`/markets/news`) `[fix, security]`- **The crypto news archive keeps its place when its storage credentials briefly drop**: Every hour a scheduled job copies the latest crypto news into the permanent archive behind the markets pages, so that a story stays readable at its own link long after it has rotated out of the live feed. Before writing anything, the job asks Google Cloud for the credentials it needs. That request can come back empty for a moment, usually because the credential service itself is briefly unreachable rather than because anything is actually misconfigured, and the job treated that as a crash: it returned an unexplained internal error and raised an alert, which read as though the archive were broken. It now recognises a missing credential for what it is, records the hour as a skipped run with the reason stated plainly, and picks up again on the next pass. Nothing is lost when that happens, because each article is filed under a fingerprint of its own link, so the next successful run archives exactly the stories the skipped one would have and never files a duplicate. A genuine storage failure is still reported loudly. `[fix, infra]`@@ -72,6 +79,7 @@ Public history for [three.ws](https://three.ws), newest first. New pages come fr- **The link checker now covers the docs, blog, and chat pages it had been skipping, and three dead links are fixed**: Every build runs a link audit that walks the site and fails if any link points at a page that does not exist. It was only ever reading three directories, and the docs reader, the blog, and the chat app live in three more that the router serves just as publicly. So those pages shipped unchecked while the audit still reported a clean zero, which is the worst way for a safety check to fail: green with the gap intact. The audit now walks every directory the router can serve a page out of, and a new test derives that list from the routing table itself, so the next section added to the site cannot quietly slip past it. Two smaller blind spots went with it: a code sample printed on a documentation page was being read as if it were a real link on that page, and a link built up in JavaScript from several pieces was being reported as a dead one. Three genuinely dead links found along the way are fixed, and the audit is now part of the pre-build gate rather than something someone has to remember to run. `[fix, infra]`- **The machine-readable API spec stops telling agents they need a login for a lane they can just pay for**: Agents and crawlers read /openapi.json to work out how to call this platform without a human present, and two of its entries were sending them the wrong way. The MCP endpoint was listed as requiring a bearer token or an API key, when in fact an unauthenticated call answers with a payment challenge: an agent can settle in USDC and get to work with no key, no token, and no account at all, and a credential is accepted but never required. Any agent that trusted the spec went off to obtain a login it did not need. The spec also described only the paid half of the Forge generation lane, hiding the free price lookup on the same address that lets an agent read the per-tier pricing and the accepted request shape before it commits any money. Both are corrected, the session-teardown call the MCP route already advertises in its own headers is documented, and the human API reference now matches. The sitemap, robots.txt, llms.txt, and llms-full.txt feeds were swept in the same pass and every address they publish was confirmed to resolve. (`/docs/api-reference`) `[fix, docs]`- **The market data index now shows every paid endpoint, at the price you will actually be charged**: The free front door to the paid market data API is where an agent looks before it buys anything. It only listed the seventeen endpoints in the core family, so the Market Heatmap and Market Mood endpoints were invisible to anyone discovering the bundle through it, and it quoted each endpoint's default price rather than the price the endpoint currently charges, so an operator price change left the index advertising a stale number. The index now builds itself from both sources, quotes the live price, and prints sub-cent prices in full instead of rounding them down to zero. Separately, a coin the upstream had not ranked yet could appear in the Market Heatmap board carrying rank 0; it now takes its position on the board instead. (`/docs/market-data-api`) `[fix, improvement]`+- **The marketplace, the directory and the home page open with a lot less work behind them**: A Lighthouse sweep of eight pages against the live site found four costs that every visitor was paying for content they could not see. The marketplace was laying out its entire catalogue on first paint, a grid measured at over 31,000 pixels tall of which one screenful is visible, and the directory did the same at smaller scale; both now skip the offscreen rows until you scroll near them. The analytics library was loading in the page head on every route, which cost around a second of browser time before anything was interactive, and now waits for the first idle moment instead, with nothing dropped in between. The embeddable avatar bundle on our CDN was shipping unminified because of a build-tool rule meant for packages that get re-bundled, which does not apply to a file browsers download directly; it is now 2.77 MB instead of 4.24 MB, and 792 KB instead of 1.03 MB over the wire. And the thumbnails in both grids now decode off the critical path. The rules behind all four, and how to measure before changing any of them, are written up so the next change does not quietly undo one. (`/marketplace`) `[improvement, infra]`- **The MCP tool catalog now lists three tools it was hiding, and stops calling the x402 Bazaar free**: The public tool catalog at /mcp-tools is generated from the code so it cannot drift, which is exactly why two wrong things in it went unnoticed. The free 3D Studio connector builds its tool list from two source files and the generator only ever read one of them, so the three persona tools that give a generated body a name, bring it back in a later session, and make it lip-sync a reply were served by the endpoint but missing from the catalog, from the machine-readable /mcp-catalog.json an agent fetches to discover what it can call, and from the two build gates that hold every published tool to its declared contract and its declared safety annotations. All three are now covered by both gates and appear in the catalog. Separately, the x402 Bazaar was labelled as needing no account and no payment; a real search call with neither returns a payment request, so the catalog now says OAuth or x402, matching the server, its documentation, and the fleet directory at /.well-known/mcp.json. (`/mcp-tools`) `[fix, docs, sdk]`- **The one-tag Shopify shopping assistant installs again, and three guides stop teaching things the code never did**: The install snippet for the Shopify shopping assistant carried a module attribute on its script tag, which quietly made it unloadable on every store: a module script is fetched cross-origin under stricter rules than a plain one, and the file it points at does not send the headers those rules require, so the widget simply never appeared. The tag is corrected everywhere it is published, so pasting it into a theme works on the first try. The self-hosted NVIDIA TRELLIS lane had a similar mismatch underneath: when no dedicated address was set it fell back to the address of our own generation worker, which answers a different protocol, so the health check reported a ready service and every generation then failed against it. That lane now reads only its own setting and says plainly when one is missing. Three tutorials were rewritten against the code they describe: the ERC-8004 reputation guide was calling three contract functions and one agent tool that do not exist on the live registry and now documents the real interface, the Solana reputation guide pointed at paid lookups that were never shipped and now shows the free ones that are, and the Tour Builder page was missing a heading for screen readers. (`/tutorials/shopify-shopping-assistant`) `[fix, docs]`- **The Oracle live action stream no longer goes quiet after a bad reconnect**: The trading floor at /activity reconnects to its live feed by telling the server the timestamp it last saw. If that timestamp arrived malformed, the server opened the stream anyway and then silently failed every lookup behind it: the page stayed connected, showed no error, and never received another action for as long as it was open. It now rejects the bad timestamp up front with a clear message so the page reconnects cleanly instead of sitting on a dead feed. Three neighbouring Oracle endpoints got the same treatment: a malformed paging cursor on the activity feed, and a non-text agent or chat id sent to the follow endpoint, used to come back as an unexplained server error and now come back as a plain description of what was wrong. An unreadable result limit on an agent's public track record no longer wipes out its recent-trade list. The activity feed and the calibration report also now allow cross-site reads, matching every other public Oracle endpoint. (`/activity`) `[fix, improvement]`diff --git a/data/changelog.json b/data/changelog.jsonindex 03a627029..d9b92a79a 100644--- a/data/changelog.json+++ b/data/changelog.json@@ -1,10 +1,46 @@{"$comment": "Curated changelog entries for three.ws, the editorial layer of the public changelog. New PAGE launches are tracked automatically via the `added` field in data/pages.json; this file is for everything else holders care about: improvements, fixes, SDK releases, trading/payments capabilities, security work. scripts/build-page-index.mjs merges both sources into CHANGELOG.md, public/changelog.json, and public/changelog.xml. Newest first. Allowed tags: feature, improvement, fix, sdk, infra, docs, security. `link` is optional and must be a live three.ws path.","entries": [+ {+ "date": "2026-08-14",+ "title": "The AR docs stopped handing readers three commands the tool rejects or quietly ruins",+ "summary": "Eight pages were read line by line against the code and the live site: the agent abilities dossier, news, the changelog, AR viewing, BNB payments, Agent Sniper, earned autonomy, and the agent personality tutorial. Every command printed on them was actually run. The AR page was wrong three times, and each one cost a reader real time. Its model optimization step used a flag the glTF tool does not accept, so the command simply failed. The line beside it promised a lossless result, when that tool compresses and simplifies by default: the file it produced could not be opened by the iPhone and Android AR viewers the page is about, and nothing said so. Its troubleshooting step for exactly that failure repeated the same rejected flag, so the fix for the problem hit the problem again. All three now carry flags the tool accepts, verified by running them on a real model and reading back what came out, with a note about what the defaults do. The AR page also linked to a placeholder address rather than a page, and now points at a real one. On the earned autonomy page, the list of settings a trading arm unlocks once it has proven itself was one short, missing the size of the position it keeps riding after taking its stake back. Everything else held up, including all 4,854 links across the eight pages.",+ "tags": [+ "docs",+ "fix"+ ],+ "link": "/docs/ar"+ },+ {+ "date": "2026-08-14",+ "title": "Links inside the tutorials now go where they say, and the AWS page names the right region",+ "summary": "Every tutorial on the site is written as a markdown file that lives beside the other documentation, and the links inside it were written against that location. The page that renders them was reading those links as if they pointed at site addresses instead, so a tutorial that sent you to a related guide, or cited the exact source file a step describes, sent you to an address that does not exist. Links that reach across to another document, that point at a specific section of one, or that cite a file in the source tree are now each resolved to the right destination, with source files opening on GitHub. The scene tutorial's own list of related tutorials pointed at a secondary address that rendered them under a generic title, and now uses the real one. Separately, the AWS page claimed the platform's production deployment runs in AWS us-east-1. It does not: production runs on Google Cloud Run, and us-east-1 is where the AWS Marketplace billing and entitlement APIs are called. The page now says that instead.",+ "tags": [+ "docs",+ "fix"+ ]+ },+ {+ "date": "2026-08-14",+ "title": "The commit feed went quiet for thirteen hours, and the reason it could not restart itself is fixed",+ "summary": "Every commit that lands on the platform is posted to the holders' Telegram channel as it happens. That feed stopped at 06:45 UTC and stayed silent for the rest of the day, and the cause was in how it found its place again after each run. It looked back over only the thirty most recent commits for the last one it had announced. On a day when work lands in bursts of twenty or thirty at a time, that marker regularly fell out of view, and when it did, the feed quietly skipped to the newest commit and posted nothing at all. Every commit in the gap was passed over with no record that anything had been missed, and because each run repeated the same skip, the feed had no way back on its own. It now pages back as far as five hundred commits to find its place, and when it is behind it works forward from the oldest commit it still owes rather than jumping to the front, so a backlog is delivered in order over the following runs instead of being discarded. Anything more than three days old is still left alone, so catching up can never flood the channel with history. The 117 commits missed during today's outage are not recoverable and will not appear.",+ "tags": [+ "fix"+ ]+ },+ {+ "date": "2026-08-14",+ "title": "A dedicated channel for first-time pump.fun creator fee claims",+ "summary": "The first time a coin creator pulls their accrued fees out of the vault is an irreversible on-chain signal that they are actually engaged with what they launched, and it was only visible by polling an API. There is now a delivery lane that posts each of those first claims to its own Telegram channel as it lands, one message per claim, with the creator, the amount in SOL, and links to the transaction and the coin. It keeps its own state so switching it on never dumps a backlog into the channel, it drains a backlog in order rather than skipping ahead if it falls behind, and it recognises claims by transaction rather than by timestamp, so two claims landing in the same second both get posted instead of one being silently dropped. The feed stays quiet until a claims indexer is configured: the direct on-chain fallback samples under a second of activity on a program this busy, which is not enough to tell a first claim from a repeat one.",+ "tags": [+ "feature",+ "infra"+ ]+ },{"date": "2026-08-14","title": "The choreography docs no longer teach an example that quietly does nothing",- "summary": "A routine step can pin one specific animation clip instead of letting the gesture slot pick one, and the choreography reference showed how, with a shareable link that pinned the dance slot to the rumba clip. That example never pinned anything. Rumba is already the platform default for the dance slot, and the runtime deliberately drops a clip that matches the slot's own default, so that a shared routine still adapts to an agent that has remapped its gestures. Anyone copying the example got a step that resolved normally instead of the pinned motion they asked for, with nothing to indicate why. The example now pins a clip that is not the default, and the field reference states the drop rule outright. The selfie-to-avatar page also had a summary line quoting an older, lower head-coverage number than the measurement reported further down the same page; both now read the measured figure.",+ "summary": "A routine step can pin one specific animation clip instead of letting the gesture slot pick one, and the choreography reference showed how, with a shareable link that pinned the dance slot to the rumba clip. That example never pinned anything. Rumba is already the platform default for the dance slot, and the runtime deliberately drops a clip that matches the slot's own default, so that a shared routine still adapts to an agent that has remapped its gestures. Anyone copying the example got a step that resolved normally instead of the pinned motion they asked for, with nothing to indicate why. The example now pins a clip that is not the default, and the field reference states the drop rule outright. The selfie-to-avatar page also had a summary line quoting an older, lower head-coverage number than the measurement reported further down the same page, and the avatar-thumbnails page printed a sample coverage readout from a time when three quarters of avatars still had no preview image. Both now show figures taken from the running system.","tags": ["docs","fix"diff --git a/prompts/swarm-100/docs-batch-18.md b/prompts/swarm-100/docs-batch-18.mddeleted file mode 100644index 9480b3cce..000000000--- a/prompts/swarm-100/docs-batch-18.md+++ /dev/null@@ -1,100 +0,0 @@-# Docs audit batch 18: /docs/agent-abilities through /tutorials/agent-personality--How to run: paste this file's repo path into a fresh Claude Code chat in this-repository and say "run this work order". This file is fully self-contained:-it depends on no other prompt file anywhere. If neighboring files in-prompts/swarm-100/ are gone, that work is done; if present, ignore them.-Every claim below rots; step 0 re-measures, and what you measure wins.--## Operating clause (binding)--- Read CLAUDE.md first. Its rules override everything, including this file.-- Finish 100% in this session. Never end the turn with a question, an option- list, or an unexecuted plan. A judgment call goes in one line of the final- report; it never becomes a question that halts work.-- The only permitted stops are the CLAUDE.md stop-and-ask gates: spending real- funds or any irreversible on-chain write, git push or a production deploy,- committing content that references a crypto project other than $THREE, and- destroying unrecoverable data.-- No mocks, no fake data, no placeholder stubs, no unfinished-work markers, no- commented-out code. Real APIs and real integrations only.-- The em-dash and en-dash characters are banned in everything you write.-- Concurrent agents share this worktree: stage explicit paths only (never a- bare add-everything), and commit finished work promptly.-- Before committing, run: npm run check:rules -- --paths <files you touched>.- It must exit 0.--## Step 0: re-derive the current state--1. Confirm each path below still exists in data/pages.json; a path that was- removed there is out of scope (note it in the report, skip it).-2. Start (or reuse) the local dev server: npm run dev (vite, port 3000).-3. Probe production for each path:-- for p in /docs/agent-abilities /news /changelog /docs/ar /docs/bnb-payments /docs/agent-sniper /docs/sniper-autonomy /tutorials/agent-personality; do printf '%s ' "$p"; curl -so /dev/null -w '%{http_code}\n' "https://three.ws$p"; done--## Task: audit this batch of docs pages, then fix everything found--Pages in this batch:--- /docs/agent-abilities ("Docs · What Agents Can Do: The Complete Dossier")-- /news ("News and Announcements")-- /changelog ("Changelog")-- /docs/ar ("Docs: AR Viewing")-- /docs/bnb-payments ("Docs: BNB Payments")-- /docs/agent-sniper ("Agent Sniper: Autonomous On-Chain Trading")-- /docs/sniper-autonomy ("Earned Autonomy: Freedom the Sniper Fleet Has to Trade For")-- /tutorials/agent-personality ("Tutorial: Agent Personality")--For every page, in a real headless browser against localhost:--1. HTTP 200, zero console errors, zero unhandled failed requests.-2. Every link on the page resolves (internal paths exist in the route table;- anchors have targets). Fix or remove dead ones.-3. Content accuracy: the page documents current behavior. Spot-check its claims against the code it describes; where it names a command, run the command; where it shows a code sample, the sample must actually run. Fix stale claims in the same change.-4. Title tag and meta description present and specific.-5. Fix defects at their root and re-run until clean.--## Definition of done--- [ ] Every page in the batch loads locally with HTTP 200 and zero console- errors.-- [ ] Zero dead links across the batch.-- [ ] Every named command and code sample on these pages verified runnable, or corrected.-- [ ] Title and meta description verified on every page.-- [ ] npm run check:rules -- --paths <touched files> exits 0; npm test passes- if you touched code.-- [ ] Production status per page recorded in the report.--## Never blocked--| Blocker | Resolution (act, do not ask) |-|---|---|-| Missing env var or credential | Check .env and .env.local, then the Cloud Run service env, then Secret Manager (the CLAUDE.md self-unblock playbook has the exact commands). If it truly exists nowhere, wire the code fully behind the env var, prove the wiring with a dry run, and name the single missing var in the report. |-| gcloud not on PATH | Run: export PATH="$HOME/google-cloud-sdk/bin:$PATH" |-| Dev server port 3000 busy | Another agent may be serving this repo; probe it and reuse if so. Otherwise start your own on a free port: npx vite --port 3101 |-| Playwright browser missing | npx playwright install chromium |-| A surface needs a signed-in user | Register a fresh account through the real /register flow against the real API and use it. Never mock the session. |-| A defect sits in code you did not touch | Fix it if it blocks a Definition of done line (root cause it, never mask it). Otherwise note it in the report and continue. |-| An unrelated test is red | Same rule. Never pipe npm test through tail; it masks exit codes. |--## Close out (required)--1. Verify every Definition of done line with the actual command output in- front of you. Never claim a line you did not verify.-2. Commit with explicit paths and a subject that describes the diff (house- style: type(scope): what changed and why a reader cares).-3. Delete this prompt file in that same commit:-- git rm prompts/swarm-100/docs-batch-18.md-- The shrinking directory is this campaign's only progress ledger; there is- no progress log to update.-4. Final report: what step 0 measured, what changed (file list), evidence per- Definition of done line, and any one-line judgment calls. No trailing- questions, no unexecuted plans.--If a line genuinely cannot pass inside this session (an external party must-respond, or an owner-gated action is the final step), finish everything else,-leave this file in place, and state exactly which line remains and who owns-it. Never delete this file on a partial.diff --git a/public/changelog.json b/public/changelog.jsonindex d4e0c99b3..e45774a21 100644--- a/public/changelog.json+++ b/public/changelog.json@@ -1,5 +1,5 @@{- "generated_at": "2026-08-14T19:37:07.682Z",+ "generated_at": "2026-08-14T19:53:55.347Z","generated_by": "scripts/build-page-index.mjs from data/pages.json + data/changelog.json","site": {"name": "three.ws",@@ -36,6 +36,16 @@"launch"]},+ {+ "date": "2026-08-14",+ "type": "launch",+ "title": "Docs · pump.fun first-claims channel",+ "summary": "A live Telegram feed of first-time pump.fun creator fee claims: how the delivery lane scans, dedupes by transaction signature, and drains a backlog in order, plus the configuration it needs and why the direct on-chain fallback cannot serve the feed.",+ "link": "/docs/pump-claims-channel",+ "tags": [+ "launch"+ ]+ },{"date": "2026-08-14","type": "launch",@@ -46,6 +56,16 @@"launch"]},+ {+ "date": "2026-08-14",+ "type": "launch",+ "title": "Docs · Site load performance",+ "summary": "The load-cost rules behind every page that is not /play, each with the Lighthouse measurement that produced it: catalogue grids skip the rows nobody can see, analytics waits for the first idle moment, the CDN bundle ships minified, below-the-fold sections load below the fold, and thumbnails decode off the critical path.",+ "link": "/docs/site-performance",+ "tags": [+ "launch"+ ]+ },{"date": "2026-08-14","type": "launch",@@ -99,6 +119,17 @@"infra"]},+ {+ "date": "2026-08-14",+ "type": "update",+ "title": "A dedicated channel for first-time pump.fun creator fee claims",+ "summary": "The first time a coin creator pulls their accrued fees out of the vault is an irreversible on-chain signal that they are actually engaged with what they launched, and it was only visible by polling an API. There is now a delivery lane that posts each of those first claims to its own Telegram channel as it lands, one message per claim, with the creator, the amount in SOL, and links to the transaction and the coin. It keeps its own state so switching it on never dumps a backlog into the channel, it drains a backlog in order rather than skipping ahead if it falls behind, and it recognises claims by transaction rather than by timestamp, so two claims landing in the same second both get posted instead of one being silently dropped. The feed stays quiet until a claims indexer is configured: the direct on-chain fallback samples under a second of activity on a program this busy, which is not enough to tell a first claim from a repeat one.",+ "link": null,+ "tags": [+ "feature",+ "infra"+ ]+ },{"date": "2026-08-14","type": "update",@@ -392,6 +423,17 @@"fix"]},+ {+ "date": "2026-08-14",+ "type": "update",+ "title": "Links inside the tutorials now go where they say, and the AWS page names the right region",+ "summary": "Every tutorial on the site is written as a markdown file that lives beside the other documentation, and the links inside it were written against that location. The page that renders them was reading those links as if they pointed at site addresses instead, so a tutorial that sent you to a related guide, or cited the exact source file a step describes, sent you to an address that does not exist. Links that reach across to another document, that point at a specific section of one, or that cite a file in the source tree are now each resolved to the right destination, with source files opening on GitHub. The scene tutorial's own list of related tutorials pointed at a secondary address that rendered them under a generic title, and now uses the real one. Separately, the AWS page claimed the platform's production deployment runs in AWS us-east-1. It does not: production runs on Google Cloud Run, and us-east-1 is where the AWS Marketplace billing and entitlement APIs are called. The page now says that instead.",+ "link": null,+ "tags": [+ "docs",+ "fix"+ ]+ },{"date": "2026-08-14","type": "update",@@ -552,6 +594,17 @@"docs"]},+ {+ "date": "2026-08-14",+ "type": "update",+ "title": "The AR docs stopped handing readers three commands the tool rejects or quietly ruins",+ "summary": "Eight pages were read line by line against the code and the live site: the agent abilities dossier, news, the changelog, AR viewing, BNB payments, Agent Sniper, earned autonomy, and the agent personality tutorial. Every command printed on them was actually run. The AR page was wrong three times, and each one cost a reader real time. Its model optimization step used a flag the glTF tool does not accept, so the command simply failed. The line beside it promised a lossless result, when that tool compresses and simplifies by default: the file it produced could not be opened by the iPhone and Android AR viewers the page is about, and nothing said so. Its troubleshooting step for exactly that failure repeated the same rejected flag, so the fix for the problem hit the problem again. All three now carry flags the tool accepts, verified by running them on a real model and reading back what came out, with a note about what the defaults do. The AR page also linked to a placeholder address rather than a page, and now points at a real one. On the earned autonomy page, the list of settings a trading arm unlocks once it has proven itself was one short, missing the size of the position it keeps riding after taking its stake back. Everything else held up, including all 4,854 links across the eight pages.",+ "link": "/docs/ar",+ "tags": [+ "docs",+ "fix"+ ]+ },{"date": "2026-08-14","type": "update",@@ -585,6 +638,17 @@"docs"]},+ {+ "date": "2026-08-14",+ "type": "update",+ "title": "The choreography docs no longer teach an example that quietly does nothing",+ "summary": "A routine step can pin one specific animation clip instead of letting the gesture slot pick one, and the choreography reference showed how, with a shareable link that pinned the dance slot to the rumba clip. That example never pinned anything. Rumba is already the platform default for the dance slot, and the runtime deliberately drops a clip that matches the slot's own default, so that a shared routine still adapts to an agent that has remapped its gestures. Anyone copying the example got a step that resolved normally instead of the pinned motion they asked for, with nothing to indicate why. The example now pins a clip that is not the default, and the field reference states the drop rule outright. The selfie-to-avatar page also had a summary line quoting an older, lower head-coverage number than the measurement reported further down the same page, and the avatar-thumbnails page printed a sample coverage readout from a time when three quarters of avatars still had no preview image. Both now show figures taken from the running system.",+ "link": "/docs/choreography",+ "tags": [+ "docs",+ "fix"+ ]+ },{"date": "2026-08-14","type": "update",@@ -596,6 +660,16 @@"infra"]},+ {+ "date": "2026-08-14",+ "type": "update",+ "title": "The commit feed went quiet for thirteen hours, and the reason it could not restart itself is fixed",+ "summary": "Every commit that lands on the platform is posted to the holders' Telegram channel as it happens. That feed stopped at 06:45 UTC and stayed silent for the rest of the day, and the cause was in how it found its place again after each run. It looked back over only the thirty most recent commits for the last one it had announced. On a day when work lands in bursts of twenty or thirty at a time, that marker regularly fell out of view, and when it did, the feed quietly skipped to the newest commit and posted nothing at all. Every commit in the gap was passed over with no record that anything had been missed, and because each run repeated the same skip, the feed had no way back on its own. It now pages back as far as five hundred commits to find its place, and when it is behind it works forward from the oldest commit it still owes rather than jumping to the front, so a backlog is delivered in order over the following runs instead of being discarded. Anything more than three days old is still left alone, so catching up can never flood the channel with history. The 117 commits missed during today's outage are not recoverable and will not appear.",+ "link": null,+ "tags": [+ "fix"+ ]+ },{"date": "2026-08-14","type": "update",@@ -719,6 +793,17 @@"improvement"]},+ {+ "date": "2026-08-14",+ "type": "update",+ "title": "The marketplace, the directory and the home page open with a lot less work behind them",+ "summary": "A Lighthouse sweep of eight pages against the live site found four costs that every visitor was paying for content they could not see. The marketplace was laying out its entire catalogue on first paint, a grid measured at over 31,000 pixels tall of which one screenful is visible, and the directory did the same at smaller scale; both now skip the offscreen rows until you scroll near them. The analytics library was loading in the page head on every route, which cost around a second of browser time before anything was interactive, and now waits for the first idle moment instead, with nothing dropped in between. The embeddable avatar bundle on our CDN was shipping unminified because of a build-tool rule meant for packages that get re-bundled, which does not apply to a file browsers download directly; it is now 2.77 MB instead of 4.24 MB, and 792 KB instead of 1.03 MB over the wire. And the thumbnails in both grids now decode off the critical path. The rules behind all four, and how to measure before changing any of them, are written up so the next change does not quietly undo one.",+ "link": "/marketplace",+ "tags": [+ "improvement",+ "infra"+ ]+ },{"date": "2026-08-14","type": "update",diff --git a/public/changelog.xml b/public/changelog.xmlindex ce49ec651..3b71e4241 100644--- a/public/changelog.xml+++ b/public/changelog.xml@@ -30,6 +30,14 @@<category>launch</category><description>How three.ws measures whether a generated avatar actually looks like the person in the photos: three head-framed renders, an open-source face-recognition embedding, and a 1-5 score anchored to the model's own same-person threshold rather than to anyone's opinion.</description></item>+ <item>+ <title>Docs · pump.fun first-claims channel</title>+ <link>https://three.ws/docs/pump-claims-channel</link>+ <guid isPermaLink="false">2026-08-14:Docs · pump.fun first-claims channel</guid>+ <pubDate>Fri, 14 Aug 2026 12:00:00 GMT</pubDate>+ <category>launch</category>+ <description>A live Telegram feed of first-time pump.fun creator fee claims: how the delivery lane scans, dedupes by transaction signature, and drains a backlog in order, plus the configuration it needs and why the direct on-chain fallback cannot serve the feed.</description>+ </item><item><title>Docs · Recurring payments</title><link>https://three.ws/docs/recurring-payments</link>@@ -38,6 +46,14 @@<category>launch</category><description>Sign one permission, then a fixed USDC amount moves every day or week, or gets swapped into a token on a schedule. The lifecycle, the API, what every kind of failed charge does to a schedule, and the operator config settlement needs.</description></item>+ <item>+ <title>Docs · Site load performance</title>+ <link>https://three.ws/docs/site-performance</link>+ <guid isPermaLink="false">2026-08-14:Docs · Site load performance</guid>+ <pubDate>Fri, 14 Aug 2026 12:00:00 GMT</pubDate>+ <category>launch</category>+ <description>The load-cost rules behind every page that is not /play, each with the Lighthouse measurement that produced it: catalogue grids skip the rows nobody can see, analytics waits for the first idle moment, the CDN bundle ships minified, below-the-fold sections load below the fold, and thumbnails decode off the critical path.</description>+ </item><item><title>Docs · The Agent Index</title><link>https://three.ws/docs/agent-index</link>@@ -78,6 +94,14 @@<category>fix</category><description>Two of the scheduled jobs behind the trading fleet could fail without anything looking wrong. The job that reallocates budget across strategies talks to the database over a socket, and when that socket never opened the failure arrived as an object carrying no message at all. The job caught it, answered with an empty error string, and the scheduler recorded a success, so the health record stayed green while the loop did nothing. Failures of that kind are now recognised as what they are, a database that could not be reached, so the run is marked unhealthy and retried on the next tick instead of being logged as fine. Every page and endpoint that reads from the database inherits the same correction: this class of outage now returns a retryable service-unavailable response rather than an internal error. The fleet watchdog also gained a seventh check, over the calibration pass that measures whether a high-conviction coin actually wins as often as its score claims, so that loop can no longer go quiet unnoticed.</description></item>+ <item>+ <title>A dedicated channel for first-time pump.fun creator fee claims</title>+ <link>https://three.ws/changelog</link>+ <guid isPermaLink="false">2026-08-14:A dedicated channel for first-time pump.fun creator fee claims</guid>+ <pubDate>Fri, 14 Aug 2026 12:00:00 GMT</pubDate>+ <category>feature</category>+ <description>The first time a coin creator pulls their accrued fees out of the vault is an irreversible on-chain signal that they are actually engaged with what they launched, and it was only visible by polling an API. There is now a delivery lane that posts each of those first claims to its own Telegram channel as it lands, one message per claim, with the creator, the amount in SOL, and links to the transaction and the coin. It keeps its own state so switching it on never dumps a backlog into the channel, it drains a backlog in order rather than skipping ahead if it falls behind, and it recognises claims by transaction rather than by timestamp, so two claims landing in the same second both get posted instead of one being silently dropped. The feed stays quiet until a claims indexer is configured: the direct on-chain fallback samples under a second of activity on a program this busy, which is not enough to tell a first claim from a repeat one.</description>+ </item><item><title>A failed database migration can no longer freeze the schema in silence</title><link>https://three.ws/changelog</link>@@ -294,6 +318,14 @@<category>fix</category><description>Two accessibility defects on high-traffic pages are fixed. The shared navigation injects a Skip to content link as the first thing you reach with the Tab key, so keyboard and screen-reader users can jump straight to the page body instead of walking the whole header. It only checked for its own copy of that link before injecting, but roughly three dozen pages write the anchor into their own markup under a different class, so those pages ended up with two identical skip links and you had to tab past the same control twice to get anywhere. The check now recognises every variant, so there is exactly one skip link per page. On the homepage, the hint under the interactive dragon that reads move cursor, hold to breathe fire was drawn in white at 18 percent opacity over a near-black panel, about 1.6 to 1 against the background where the accessibility standard asks for at least 4.5 to 1; anyone in bright light or with reduced contrast sensitivity simply could not see the instruction for the thing it was explaining. It now sits at 5.7 to 1 and stays legible without changing the look of the scene. The Genesis and Walk pages also had a skip link with no styling rule attached, which left it as stray visible text above the header; both now use the shared style and hide until focused like every other page.</description></item>+ <item>+ <title>Links inside the tutorials now go where they say, and the AWS page names the right region</title>+ <link>https://three.ws/changelog</link>+ <guid isPermaLink="false">2026-08-14:Links inside the tutorials now go where they say, and the AWS page names the right region</guid>+ <pubDate>Fri, 14 Aug 2026 12:00:00 GMT</pubDate>+ <category>docs</category>+ <description>Every tutorial on the site is written as a markdown file that lives beside the other documentation, and the links inside it were written against that location. The page that renders them was reading those links as if they pointed at site addresses instead, so a tutorial that sent you to a related guide, or cited the exact source file a step describes, sent you to an address that does not exist. Links that reach across to another document, that point at a specific section of one, or that cite a file in the source tree are now each resolved to the right destination, with source files opening on GitHub. The scene tutorial's own list of related tutorials pointed at a secondary address that rendered them under a generic title, and now uses the real one. Separately, the AWS page claimed the platform's production deployment runs in AWS us-east-1. It does not: production runs on Google Cloud Run, and us-east-1 is where the AWS Marketplace billing and entitlement APIs are called. The page now says that instead.</description>+ </item><item><title>Oracle pages now say the data is briefly unavailable instead of showing you a blank market</title><link>https://three.ws/changelog</link>@@ -414,6 +446,14 @@<category>fix</category><description>The demos index has always carried a card for the agent interaction lab, the set of small pages where an avatar high-fives you, climbs the page title, bounces on a trampoline, or falls asleep when you leave it alone. Clicking that card gave you a 404. The lab hub lives in a folder rather than as a single page, and the rule that turns a demo name into a page file was sending the request to a file that never existed, so the one link into the lab was the only link on the index that could not be followed. The sub-demos were fine the whole time; only the door was missing. It is fixed, and the demo route map now records the whole lab, including the lost-avatar 404 page that has its own card on the index but was never listed.</description></item>+ <item>+ <title>The AR docs stopped handing readers three commands the tool rejects or quietly ruins</title>+ <link>https://three.ws/docs/ar</link>+ <guid isPermaLink="false">2026-08-14:The AR docs stopped handing readers three commands the tool rejects or quietly ruins</guid>+ <pubDate>Fri, 14 Aug 2026 12:00:00 GMT</pubDate>+ <category>docs</category>+ <description>Eight pages were read line by line against the code and the live site: the agent abilities dossier, news, the changelog, AR viewing, BNB payments, Agent Sniper, earned autonomy, and the agent personality tutorial. Every command printed on them was actually run. The AR page was wrong three times, and each one cost a reader real time. Its model optimization step used a flag the glTF tool does not accept, so the command simply failed. The line beside it promised a lossless result, when that tool compresses and simplifies by default: the file it produced could not be opened by the iPhone and Android AR viewers the page is about, and nothing said so. Its troubleshooting step for exactly that failure repeated the same rejected flag, so the fix for the problem hit the problem again. All three now carry flags the tool accepts, verified by running them on a real model and reading back what came out, with a note about what the defaults do. The AR page also linked to a placeholder address rather than a page, and now points at a real one. On the earned autonomy page, the list of settings a trading arm unlocks once it has proven itself was one short, missing the size of the position it keeps riding after taking its stake back. Everything else held up, including all 4,854 links across the eight pages.</description>+ </item><item><title>The Arena now runs a daily competition on its own, instead of waiting for someone to host one</title><link>https://three.ws/arena</link>@@ -438,6 +478,14 @@<category>fix</category><description>Paste a contract address into the CA to x402 tool and it generates the copy-paste code for calling the paid endpoint it just built for you. The cURL tab's second step told you to pay by running npx x402-fetch against the endpoint. That never worked: the published x402-fetch package is a library, not a command line tool, so anyone who copied it got an npm error about not being able to determine an executable to run, on their first attempt to pay us. The tab now shows the real two-step HTTP flow instead, taking the 402 and then replaying the same call with the payment header, and says outright that the Node and Agent tabs are what sign that header for you. Both commands were run end to end against the live endpoint before shipping. The docs page carries the same correction, and the vault and master wallet docs pages got proper search descriptions in place of ones that quoted raw filenames at readers.</description></item>+ <item>+ <title>The choreography docs no longer teach an example that quietly does nothing</title>+ <link>https://three.ws/docs/choreography</link>+ <guid isPermaLink="false">2026-08-14:The choreography docs no longer teach an example that quietly does nothing</guid>+ <pubDate>Fri, 14 Aug 2026 12:00:00 GMT</pubDate>+ <category>docs</category>+ <description>A routine step can pin one specific animation clip instead of letting the gesture slot pick one, and the choreography reference showed how, with a shareable link that pinned the dance slot to the rumba clip. That example never pinned anything. Rumba is already the platform default for the dance slot, and the runtime deliberately drops a clip that matches the slot's own default, so that a shared routine still adapts to an agent that has remapped its gestures. Anyone copying the example got a step that resolved normally instead of the pinned motion they asked for, with nothing to indicate why. The example now pins a clip that is not the default, and the field reference states the drop rule outright. The selfie-to-avatar page also had a summary line quoting an older, lower head-coverage number than the measurement reported further down the same page, and the avatar-thumbnails page printed a sample coverage readout from a time when three quarters of avatars still had no preview image. Both now show figures taken from the running system.</description>+ </item><item><title>The commit and changelog feeds can no longer repeat a post they already sent</title><link>https://three.ws/changelog</link>@@ -446,6 +494,14 @@<category>fix</category><description>The jobs that deliver commits and changelog entries to the holders' Telegram channel take a lock so two overlapping runs cannot announce the same thing twice, and that lock lasts four minutes. But the calls they make to GitHub and Telegram had no time limit of their own, so a single stalled call could outlive the lock it was holding, let the next run pick the lock up, and post the same commits again. Every outbound call in those jobs is now bounded well inside the lock's lifetime, so the no-duplicates guarantee holds even when an upstream service hangs. The same missing limit is fixed across four other scheduled jobs: the wardrobe catalog audit, the garment job sweep, and the weekly quality benchmark, whose own run budget was previously advisory because it was only ever checked between requests and never during one. A stalled status check no longer fails a generation that is still healthy, and a slow download is no longer reported as a corrupted garment.</description></item>+ <item>+ <title>The commit feed went quiet for thirteen hours, and the reason it could not restart itself is fixed</title>+ <link>https://three.ws/changelog</link>+ <guid isPermaLink="false">2026-08-14:The commit feed went quiet for thirteen hours, and the reason it could not restart itself is fixed</guid>+ <pubDate>Fri, 14 Aug 2026 12:00:00 GMT</pubDate>+ <category>fix</category>+ <description>Every commit that lands on the platform is posted to the holders' Telegram channel as it happens. That feed stopped at 06:45 UTC and stayed silent for the rest of the day, and the cause was in how it found its place again after each run. It looked back over only the thirty most recent commits for the last one it had announced. On a day when work lands in bursts of twenty or thirty at a time, that marker regularly fell out of view, and when it did, the feed quietly skipped to the newest commit and posted nothing at all. Every commit in the gap was passed over with no record that anything had been missed, and because each run repeated the same skip, the feed had no way back on its own. It now pages back as far as five hundred commits to find its place, and when it is behind it works forward from the oldest commit it still owes rather than jumping to the front, so a backlog is delivered in order over the following runs instead of being discarded. Anything more than three days old is still left alone, so catching up can never flood the channel with history. The 117 commits missed during today's outage are not recoverable and will not appear.</description>+ </item><item><title>The contracts that hold agent earnings, reputation stakes and skill licenses are now audit-ready</title><link>https://three.ws/changelog</link>@@ -534,6 +590,14 @@<category>fix</category><description>The free front door to the paid market data API is where an agent looks before it buys anything. It only listed the seventeen endpoints in the core family, so the Market Heatmap and Market Mood endpoints were invisible to anyone discovering the bundle through it, and it quoted each endpoint's default price rather than the price the endpoint currently charges, so an operator price change left the index advertising a stale number. The index now builds itself from both sources, quotes the live price, and prints sub-cent prices in full instead of rounding them down to zero. Separately, a coin the upstream had not ranked yet could appear in the Market Heatmap board carrying rank 0; it now takes its position on the board instead.</description></item>+ <item>+ <title>The marketplace, the directory and the home page open with a lot less work behind them</title>+ <link>https://three.ws/marketplace</link>+ <guid isPermaLink="false">2026-08-14:The marketplace, the directory and the home page open with a lot less work behind them</guid>+ <pubDate>Fri, 14 Aug 2026 12:00:00 GMT</pubDate>+ <category>improvement</category>+ <description>A Lighthouse sweep of eight pages against the live site found four costs that every visitor was paying for content they could not see. The marketplace was laying out its entire catalogue on first paint, a grid measured at over 31,000 pixels tall of which one screenful is visible, and the directory did the same at smaller scale; both now skip the offscreen rows until you scroll near them. The analytics library was loading in the page head on every route, which cost around a second of browser time before anything was interactive, and now waits for the first idle moment instead, with nothing dropped in between. The embeddable avatar bundle on our CDN was shipping unminified because of a build-tool rule meant for packages that get re-bundled, which does not apply to a file browsers download directly; it is now 2.77 MB instead of 4.24 MB, and 792 KB instead of 1.03 MB over the wire. And the thumbnails in both grids now decode off the critical path. The rules behind all four, and how to measure before changing any of them, are written up so the next change does not quietly undo one.</description>+ </item><item><title>The MCP tool catalog now lists three tools it was hiding, and stops calling the x402 Bazaar free</title><link>https://three.ws/mcp-tools</link>diff --git a/public/features.json b/public/features.jsonindex 51c2e2d9c..88a7eda0f 100644--- a/public/features.json+++ b/public/features.json@@ -1,5 +1,5 @@{- "generated_at": "2026-08-14T19:37:07.244Z",+ "generated_at": "2026-08-14T19:53:54.807Z","generated_by": "scripts/build-page-index.mjs from data/pages.json","site": {"name": "three.ws",@@ -2947,6 +2947,16 @@"tags": [],"showcase": false},+ {+ "path": "/docs/site-performance",+ "title": "Docs · Site load performance",+ "description": "The load-cost rules behind every page that is not /play, each with the Lighthouse measurement that produced it: catalogue grids skip the rows nobody can see, analytics waits for the first idle moment, the CDN bundle ships minified, below-the-fold sections load below the fold, and thumbnails decode off the critical path.",+ "added": "2026-08-14",+ "auth": null,+ "indexable": true,+ "tags": [],+ "showcase": false+ },{"path": "/docs/farcaster-memory-seeding","title": "Docs · Farcaster memory seeding",@@ -2987,6 +2997,16 @@"tags": [],"showcase": false},+ {+ "path": "/docs/pump-claims-channel",+ "title": "Docs · pump.fun first-claims channel",+ "description": "A live Telegram feed of first-time pump.fun creator fee claims: how the delivery lane scans, dedupes by transaction signature, and drains a backlog in order, plus the configuration it needs and why the direct on-chain fallback cannot serve the feed.",+ "added": "2026-08-14",+ "auth": null,+ "indexable": true,+ "tags": [],+ "showcase": false+ },{"path": "/docs/agent-tokens","title": "Docs · Agent tokens",