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): publish the API hardening entries to the holder feeds
nirholas2e375ba2d570
diff --git a/CHANGELOG.md b/CHANGELOG.mdindex d686eb1e8..c1642fd9a 100644--- a/CHANGELOG.md+++ b/CHANGELOG.md@@ -7,6 +7,7 @@ Public history for [three.ws](https://three.ws), newest first. New pages come fr## 2026-08-15- **A broken paid endpoint is no longer reported as a free one**: Payment sessions let you hand an agent a budget instead of a key, and two parts of that path judged an endpoint by one rule: if it did not ask for payment, the call was free. That is only true when the endpoint actually succeeded. A paid API answering 404, 400, or 500 was being scored as a free call, and the consequences ran in the wrong direction in both places. The dry run at /pay/simulator, whose whole job is to tell you in advance whether a budget and an allowlist will carry a workload, reported a plan as feasible at a cost of nothing when in fact every endpoint in it was erroring: free calls are never priced, never allowlist-checked, and never counted, so a wall of failures looked like a clean run. And an agent spending through a session got back a success envelope saying no payment was needed for an endpoint that had just failed on it. Both now separate the two cases: a real success with no payment challenge is still free and still costs nothing, while an error status is reported as an endpoint failure carrying the status the endpoint actually returned. No session is charged either way. The simulator lists each failing endpoint by name in its problems list and no longer calls the plan feasible. (`/pay/simulator`) `[fix]`+- **A market intelligence outage no longer tells you your own API key was rejected**: The two market intelligence reads in the unified API, recent narrative intel and momentum-ranked projects, pass a key we hold to the upstream provider; a caller needs no key at all. When that provider stopped accepting our key, both reads answered "unauthorized" with the provider's own wording, so an agent reading the response concluded its own credentials were bad and retried with different ones forever, when nothing on the caller's side was wrong. Both reads now say plainly that the service is temporarily unavailable because this deployment's upstream key was rejected, which is a status clients already know to retry rather than re-authenticate against. The older market intelligence endpoints always classified this correctly; the classification now lives in one place, so the two doors cannot drift apart again, and the error tables in the API reference list every code either one can return. (`/docs/api-reference`) `[fix, docs]`- **A retried agent spend can no longer hand its transaction to somebody else, and an unsupported chain says so instead of blaming the network**: When an agent redeems a permission you granted it, it can send a retry key so a request whose answer it never saw is not paid for twice. That key was filed under the permission alone rather than under the agent that used it, and it was read before the check that the caller owns the permission at all, so a different agent replaying the same permission and key was handed the original transaction hash and receipt instead of being turned away. The key is now filed per caller: a retry only ever returns your own result, and anyone else gets the same not-found answer they should always have got. Two more fixes in the same endpoint. A permission recorded on a chain three.ws has no delegation contract on could never settle, but the attempt came back as a network error, which reads as a temporary outage worth retrying rather than the permanent chain mismatch it is; it is now named plainly and refused before it consumes any of the permission's spending allowance. And the relayer kept its own private one-address-per-chain list of blockchain endpoints, separate from the failover chain the rest of the platform uses, so it would have refused to work on the next supported chain even where the shared endpoints reach it fine. It now uses the shared list, with any operator override still taking priority. The permissions guide gained a section on how the retry key behaves and how long it lasts. (`/docs/permissions`) `[security, fix]`- **A shared link to a docs page now opens that page instead of stranding you**: Every cross-reference inside a docs page turns into a link like /docs/agent-wallets#financial-controls. Following one worked, but the URL you copied afterwards did not: on a fresh load the viewer only recognised the second half if that doc happened to be listed in the sidebar, and silently left you on the page you arrived from otherwise. Links to Financial controls, the x402 payment tutorial, and the wallet access and proof-of-custody chapters all failed that way. The viewer now asks the server whether the doc exists before giving up, so those links land where they say they will, and the browser tab takes the destination's name. Separately, /docs/3d-api was serving the 3D API landing page rather than the reference doc that the API's own discovery response, the OpenAPI license link, and the docs sidebar all point at; it now serves the reference doc, and the landing page keeps its home at /3d. (`/docs/agent-wallets`) `[fix, docs]`- **A silent outage in the agent reasoning ledger now reports itself**: Every agent decision is committed to a hash chain, and that chain head is written to Solana on a schedule so anyone can check the history was never edited. The wallet that pays for those on-chain writes ran empty, and the scheduled job could not tell that apart from the harmless case where no signing key is set up at all: it answered a healthy 200 every ten minutes while re-sending the same doomed transaction for twelve agents in a row, and not one head reached the chain. A rejected write is now reported as a rejected write, the rest of the run stops instead of repeating a failure that cannot succeed, and it raises an alert so the wallet gets refilled. Verifying a ledger already worked without the on-chain leg and still does: the recomputed hash chain is the tamper proof, the on-chain write is the independent timestamp on top of it. Separately, the once-a-minute economy heartbeat now refuses to drive production from a machine that is not production, so a developer running it locally can no longer fire thirty-six live money engines at the real site. (`/docs/reasoning-ledger`) `[fix, infra]`@@ -14,9 +15,12 @@ Public history for [three.ws](https://three.ws), newest first. New pages come fr- **Agent .sol name lookups work again, and the avatar manifest schema now answers at its own address**: The avatar-agent MCP server resolves .eth and .sol names to wallet addresses. Its .sol side went through a hosted name-lookup service that has since been retired: every request came back as "not found", which reads exactly like an unregistered domain, so a working name and a nonexistent one were indistinguishable. Solana names are now read straight from the chain, from the same name-service accounts the three.ws backend already reads, so the lookup depends on nothing but a Solana RPC. Two things were fixed alongside it. The per-lookup time budget was under five seconds, which a cold Ethereum lookup routinely exceeds, so real names were being reported as timeouts; it is now fifteen seconds and adjustable. And a lookup that genuinely fails now says why instead of quietly reporting "not found". Separately, the avatar manifest schema names https://three.ws/schema/avatar.v1.json as its own identifier, but nothing was served there, so any validator that followed the identifier got a 404. The schema is now published at that address, with a test that keeps the served copy identical to the one shipped in the package. `[fix, sdk]`- **Agent Screen is usable on a phone, and reads correctly to a screen reader**: Watching an agent work on a narrow screen was close to impossible. The row of viewer controls along the top (fit, screenshot, picture-in-picture, fullscreen, ambient, zen, and the link back to the agent) ran off the edge of the page with no way to reach them, the agent name and its live badge painted over the buttons underneath, and the two panels that open by default covered the live screen and the avatar cam completely: the whole point of the page, hidden. The control row now scrolls sideways so every control stays reachable at any width, the name keeps its space, and on a phone the page opens on a bare stage with the panels one tap away. Three fixes for anyone using a keyboard or a screen reader landed with it: the page now has a real heading and a main region for the skip link to jump to, the hidden Skip to content link stopped rendering as visible text at the top of the page, and every text field on the page gained a proper label. Finally, if the agent list fails to load for any reason other than being signed out, the page says what went wrong and offers a retry instead of telling a signed-in owner to go sign in. (`/agent-screen`) `[fix, improvement]`- **Arming an Oracle agent for live trading no longer rounds your trade size up**: The arm page clamps the risk knobs on the server so a bad number can never reach the trading loop, and those clamps have a floor: the smallest trade the loop will place is 0.001 SOL. That floor was being applied in the wrong direction. Ask to arm a live agent with a per-trade size of zero, and instead of being told no, you were armed for real spending at 0.001 SOL a trade with a half-SOL daily ceiling you never asked for. The check meant to catch this looked at the number after the clamp had already raised it, so it could never fire. Arming live now refuses any per-trade size below the floor, and any daily cap smaller than a single trade, and tells you which number to fix; nothing is armed until you do. Clamping a too-large number back down to the ceiling is unchanged, and simulate runs keep the same forgiving behaviour as before. The arm page also shows the reason a save was rejected instead of one generic line, so an ownership problem no longer reads the same as a bad trade size, and a failed Telegram test now prints what Telegram actually said. (`/oracle/arm`) `[fix, security]`+- **Asking for a voice that does not exist now tells you so, instead of quietly speaking in a different one**: The voice API accepts a provider and a voice id from the voice catalog. If the id was not one that provider actually has, nothing said so: the request was answered with audio rendered in that lane's default voice, and on the metered lanes it was billed, so a single typo in a voice id came back as a stranger's voice at full price with a success status. A well-formed id for a voice that had been removed took the opposite route and failed as a server error, which reads as our fault and tells a client to retry something that will never work. Both now return a plain rejection that names the id you sent and the catalog address to pick a real one from, and that check runs before anything is charged. Leaving the voice id out still gets you the lane default on purpose. Asking the catalog for a provider that does not exist used to return an empty list that was indistinguishable from a provider being down; it now says which providers exist. ElevenLabs voices, whose catalog grows every time you clone or add one, are still decided by ElevenLabs, but an id it rejects is reported as your request being wrong rather than ours, with any charge refunded first. Voice cloning also now asks who you are before reporting anything about how this server is configured. (`/docs/api-reference`) `[fix, improvement]`- **Avatars can move their lips again: the Audio2Face lane is back**: Every call that turns speech into facial animation had been failing with an upstream not-found error, so a talking avatar spoke with a frozen face and the one-shot say() path returned nothing usable. The hosted model id the lane pointed at had stopped resolving for our NVIDIA account. It now points at the hosted Audio2Face NIM that is actually live, verified end to end: a synthesized clip drives a 30 fps ARKit blendshape track whose jaw genuinely opens and closes with the words. The discovery helper that finds these model ids was missing the name NVIDIA publishes them under, which is why the outage looked like a configuration mystery; it now finds them. `[fix]`+- **Claim Your Wallet stopped telling real traders they had no track record**: Pasting a wallet into Claim Your Wallet could answer "not indexed yet" to a wallet with thousands of indexed pump.fun trades. The page only accepted a wallet the reputation scorer had already graded, and that scorer runs on a rollup that lags the trade indexer by design, so a wallet the platform had watched trade for days was still turned away. Any wallet with a real on-chain record now renders its full ledger and window aggregates immediately, with the reputation panel honestly reading unproven until the grade lands. Three more fixes shipped with it. Win and loss counts now cover closed positions only, so a trader still holding sixty fresh bags no longer reads as sixty losses, and a new realized profit-and-loss figure sits beside the existing net figure that charges an open bag's entry as if it were written off. Asking for devnet returned the wallet's mainnet trades; both halves of the read are now scoped to the network you asked for. And if the database is unreachable, the page now says the record is temporarily unavailable and offers a retry, instead of publishing an empty scorecard that the CDN would then serve to everyone for the next few minutes. (`/claim-wallet`) `[fix]`- **Docs pages stop opening with a wall of raw metadata, and the AWS pages say where the Marketplace listing actually stands**: Nine documentation pages are written as article drafts, so each one starts with a small block of publishing metadata: the title, where it is meant to run, its tags. That block is meant for the tooling, not for you, but the docs reader had no idea what it was and rendered it as an enormous heading at the very top of the article, above the actual writing. It is now recognised and hidden, on the page and in the Copy page text, so those articles begin where they should. Reading the same pages closely turned up claims that had drifted: several said three.ws was already live on AWS Marketplace and billing usage to your AWS invoice. The integration is built and deployed, but the listing itself has not been published yet and the pricing decision is a free subscription with no AWS metering, so the AWS pages now say exactly that and point at the page that tracks the status. The MCP article's vanity-address tool was also listed at a metered price of up to fifty cents; it is a flat five cents per grind, which is what the code has always charged. (`/docs/aws-marketplace`) `[docs, fix]`- **Eight build tutorials now match what the platform actually does**: An audit of eight tutorials found instructions that could not work. The AI-brain tutorial described a model line-up that had been superseded: it never mentioned the Claude 5 family (Sonnet 5, Opus 5, Fable 5), which are live and selectable, and it recommended Sonnet 4.6 as the default. It also sent you to screens that do not exist, telling you to add API keys under a gear icon in My Agents and to change models, sampling knobs, and streaming under a Brain menu. Provider keys actually live in the AI Provider Keys panel on your account page, the model is a field on the agent's embed policy, and streaming has no off-switch. The promised usage tab with token charts and threshold emails did not exist either; the tutorial now documents the real per-agent usage read and points at the monthly quota and per-minute rate limit, which are the controls actually enforced before your money is spent. It also now explains that a visitor can never escalate your agent onto a paid model. The custom-GLB tutorial was worse for anyone following along: it linked a Studio address that no longer resolves, and four of its commands invoked gltf-transform subcommands that have never existed, so converting an FBX, renaming animation clips, and the full Mixamo walkthrough all failed at the first step. Those are replaced with the converter the platform itself uses and a clip-renaming script that was run against a real avatar to confirm it works. Its Draco flag was wrong, its upload path pointed at the wrong page, and it still warned that an avatar without an idle clip would stand frozen in a T-pose, which stopped being true when animation retargeting became universal. Finally, the page-events tutorial listed the wrong fallback order for emote animations. (`/tutorials/connect-ai-brain`) `[docs, fix]`+- **Looking up an on-chain agent by its CAIP reference works again**: The public resolver that turns an ERC-8004 agent reference into its owner, token URI and verified agent card could not be reached at all. An agent reference ends in a slash before its token id, and the documentation told callers to percent-encode the whole reference, which turns that slash into %2F. The API refuses any address segment holding an encoded separator, because that is the same trick used to escape a path and reach files that were never meant to be served, so every correctly-formed lookup was turned away as an unknown address before it ever reached the resolver. The verification badge you can embed on your own site built its request in exactly that shape, so it never showed a result either. The resolver now accepts the reference with its slash as a normal part of the address, the badge asks for it that way, and the documentation shows the form that works. The protection against encoded separators is untouched. New tests resolve the address through the same routing the live site uses, so this cannot quietly break again. (`/docs/api-reference`) `[fix]`- **Oracle stops answering "nothing is happening" when the database is the thing that is down**: Several Oracle reads shared a habit: if the database was unreachable for a few seconds, they answered normally with an empty result. The movers panel said nothing was moving, the stats hero said no coin had ever been scored and no agent was armed, a coin's conviction sparkline came back flat, and the machine-readable signal endpoint told autonomous agents there were no plays, which is exactly the cue for an agent to stand down. Every one of those answers was then cached at the edge for a minute or two, so a blip lasting seconds kept showing a dead market long after the database was back. Those endpoints now tell the difference between a connectivity failure and an ordinary query fault: the first is reported honestly as a temporary outage with a retry hint and is never cached, the second still degrades to the empty answer it always did. The social ingest got the sharper end of the same bug: it used to reply that a batch had been accepted while reporting zero mints written, a receipt for a write that never landed. It now fails loudly so the sender retries. Shared conviction cards behave too: one built during an outage is no longer pinned to a coin's link preview for 15 minutes. (`/oracle`) `[fix, improvement]`- **Pinning an avatar to IPFS works for real avatars, not just tiny ones**: Minting an avatar from the Avatar Studio pins the model and its metadata to IPFS, and the whole file is sent through the platform so the storage credential never has to reach your browser. The endpoint said it accepted files up to 10 MB, but it was actually rejecting anything over roughly 730 KB, which is smaller than almost any real avatar. The error it gave back was a bare 'payload too large' that named no limit, so there was nothing to act on. The cause was two size limits that had never been reconciled: files are sent as text, which makes them a third larger in transit, and the transport limit was left at its default while the documented limit was set eight times higher. The limits now agree. Inline uploads are accepted up to 5 MB, which is what the request size can actually carry, and anything larger has a clear path: upload it to storage first and pin it by URL, a lane that has always accepted 50 MB. Over-sized uploads now say which ceiling was crossed. Two smaller fixes came with it: the filename you send is now the filename filed with the storage provider, instead of every avatar being stored as 'avatar.glb' and every metadata file as 'manifest.json', which made a pin list impossible to read; and a request with a genuine mistake in it, such as a file from somewhere the platform does not own, now gets an error explaining the mistake rather than a generic 'service unavailable' that hid it. The endpoint is now documented in the API reference. `[fix, docs]`- **Plan checkout: a paid upgrade can only ever settle on real mainnet, and a late payment is no longer stranded**: Paying for a plan creates a quote, you send the funds, then the server checks the chain before it upgrades you. Three things in that path were wrong. The checkout side has always refused test-network quotes in production, because test-network dollars are free from a faucet, but the confirm side still let a caller name the test network itself, and it is the confirm side that decides which chain it looks at and which dollar it accepts. Both ends now read the same setting, so a production upgrade can only ever be settled by real funds on mainnet. Second, an Ethereum-side payment has to wait twelve blocks before it counts, which meant funds sent near the end of a session could land on-chain and then be refused for arriving late, with the money already gone; that lane now honors the same one-hour grace window the Solana lane has, so an on-chain payment is always claimable. Third, confirming an Ethereum payment could reach a Solana quote and mark it expired, and asking to buy a skill from a malformed agent id answered a generic server error instead of telling you what was wrong with the request. (`/pricing`) `[fix, security]`@@ -38,6 +42,7 @@ Public history for [three.ws](https://three.ws), newest first. New pages come fr- **The plugin marketplace stops hiding a plugin when you page through it**: Browsing plugins a page at a time could show you the same plugin twice and never show another one at all. The catalogue was sorted by install count and publish date, and the built-in plugins were all published in the same instant with zero installs, so that sort left them tied and the database was free to order the tied rows differently on every request. One plugin therefore landed on two consecutive pages and one fell through the gap. The order is now decided all the way down, so paging through the catalogue returns every plugin exactly once. Searching also takes your text literally: typing a percent sign or an underscore used to act as a database wildcard and match rows it should not. Publishing got stricter at the same time, because a plugin manifest is handed to the model as a tool definition by whoever installs it: a tool name now has to be plain letters, digits, underscores, or hyphens, every tool needs a real description, and a manifest has to fit the same 64KB ceiling that importing one by URL already enforced. And pointing the importer at a repository web page instead of the raw manifest file now says exactly that, rather than showing a raw parser error. (`/marketplace`) `[fix, improvement]`- **The pump.fun trade history works again, notification settings show every category, and a bad agent id stops reading as a server crash**: Three fixes across the public MCP packages and the API behind them. Asking for a token's recent buys and sells (get_token_trades on the free pump.fun MCP server) had stopped working entirely: the platform reads several transactions in one batched call, and the free Solana providers we route through cap those batches at one call each, so they rejected the whole request and the tool returned a raw provider error instead of trades. The platform now recognises that refusal, re-sends the reads one at a time to the same provider, and remembers the cap so later calls skip the wasted attempt; every batched on-chain read benefits, including wallet and agent activity history. Separately, the notifications MCP server told agents there were six notification categories when there are seven, so an agent could never change how it hears about a finished 3D generation; all seven are now listed, along with the security notices that always reach your bell inbox. And the agent provenance log answered a non-uuid agent id with an internal error and a support reference; it now says plainly that the id must be a uuid. `[fix, sdk]`- **The trading and MCP docs now match what the code and the live API actually do**: An audit of eight documentation pages ran every command they print and checked every claim against the code behind it, then corrected what had drifted. The biggest one was in the trading tutorial: the command it gave you for verifying a fleet's fills on-chain asked for a field the leaderboard does not publish, so it returned an empty explorer link for every agent and the verification step could not be completed at all. It now uses the fields that exist, linking both legs of each trade straight to Solscan, and it skips agents whose wallet has not been provisioned yet instead of building a URL that goes nowhere. The MCP tool safety page undercounted the tools that share a name across two servers, which matters because that collision is exactly why a trust policy has to key on the namespaced id; it now lists all nine and prints a command that re-derives the list so it cannot go stale again. The MCP tool catalog itself had drifted from its source and was serving outdated descriptions for three tools, on a page whose whole promise is that it is generated from source and cannot drift; it has been regenerated. The package extraction guide had every version in its roster out of date along with its build and test columns, and still described a publisher field as missing that was set some time ago. The 10 SOL trading experiment page stated a risk rule in absolute terms that contradicted the setup script it tells you to run, which read as a bug in one of the two; the rule now carries the scope the code actually applies. (`/docs/mcp-safety`) `[docs, fix]`+- **The Tutor’s running tab and end-of-session invoice now show what you actually paid**: The Tutor charges one cent per answer and keeps an itemized tab you can reload later and close for a signed invoice. That ledger only ever knew how to store itself in a key-value service this deployment does not run, so nothing was ever written down: reopening the page never restored your history, and ending a session always produced an invoice reading "no questions" and a total of zero, however many answers you had paid for. The tab is now stored in the platform’s own database, so it survives a reload, a new browser tab, and any server restart, and the invoice lists every answer with its charge and a verification hash over the line items. Closing a session twice returns the identical invoice instead of a freshly stamped one, closing a session id that was never used no longer makes that id unusable for its real owner, and if the ledger itself is ever unreachable the page says so rather than quietly reporting that you owe nothing. Asking a question still succeeds even if the ledger write fails, because a bookkeeping outage should never cost you the answer you paid for. (`/tutor`) `[fix]`- **The vanity receipt verifier stops calling honest receipts fraudulent, and the avatar CLI takes a Solana address**: Two documented commands did not do what their page said. The command-line verifier for a provably-fair vanity address receipt scored the difficulty claim with the retired flat model, the one that treats every Base58 character as equally likely. Every receipt issued since the leading-character correction landed therefore failed its difficulty check and was reported as untrustworthy, even though the in-browser verifier at /vanity/verify passed the same receipt. The verifier now reads the model each receipt names in its own signed difficulty block, exactly as the page and the SDK do, so current receipts verify and older ones keep verifying under the model they were issued under. Separately, the Rig Doctor page showed an avatar-manifest command taking a Solana address as the owner, which the CLI rejected outright: it understood Ethereum addresses and full CAIP-10 identifiers and nothing else. Since Solana is the home chain, a bare Solana address is now accepted and recorded as solana:mainnet-beta. Alongside those, eight tutorial and reference pages were audited against the code they describe: the Vercel self-hosting guide had a stale copy of the config file and pointed at a migrations folder that has moved, the MCP guide listed three placeholder OAuth scopes where the live server publishes twelve real per-capability ones, and the Avatar CLI page pinned a version number that npm has never had. (`/docs/rig-doctor`) `[fix, docs, sdk]`- **The vanity SDK now uses the payment wallet you hand it, and the in-chat 3D avatar keeps its tuned lighting**: Three fixes in the developer packages. The @three-ws/vanity SDK documented a fetch option on its paid address-grinding lane but silently ignored it, so a developer who passed a payment-enabled fetch got a payment-required error back instead of the keypair they paid for; that option now works, as do a custom API host and key. The avatar MCP server's in-chat 3D viewer had been hand-tuned with softer shadows, brighter exposure and a neutral environment light, but the script that rebuilds it before every release still produced the old flat version, so the next release would have quietly undone the tuning; the script now produces the tuned viewer, and the same settings apply to the fallback view for chat apps that cannot run the interactive one. The tool SDK's quick-start example pointed at an API that does not exist and skipped the wrapper an MCP server needs to return anything at all, so following it produced a tool that answered with nothing; it now runs end to end against a real free three.ws endpoint. `[fix, sdk, docs]`- **Three docs pages stop pointing readers at a dead page, a missing section, and a snippet that does nothing**: The Vanity page told you to grind a long address yourself in the browser at /vanity, but that path has never existed and answers 404; the in-browser grinder lives at /vanity-wallet, and both links on the page now go there. In Troubleshooting, the fix for a Quick Look model that opens and instantly closes linked to a model-optimization section that is not on the AR page, so the link dropped you at the top with nothing to read; it now opens the section that actually covers it. The same page told you to add a responsive attribute to size an inline embed, with a snippet that set the width in CSS. Neither half worked: responsive sizing is already on by default, and the element reads the width attribute rather than the computed style, so the documented snippet never produced the 3:4 ratio it promised. The instructions now show the form that does work, and name the two things that switch it off. (`/docs/troubleshooting`) `[fix, docs]`diff --git a/public/changelog.json b/public/changelog.jsonindex e1255bb62..8e3ec50bb 100644--- a/public/changelog.json+++ b/public/changelog.json@@ -1,5 +1,5 @@{- "generated_at": "2026-08-15T15:04:09.742Z",+ "generated_at": "2026-08-15T15:11:10.865Z","generated_by": "scripts/build-page-index.mjs from data/pages.json + data/changelog.json","site": {"name": "three.ws",@@ -16,6 +16,17 @@"fix"]},+ {+ "date": "2026-08-15",+ "type": "update",+ "title": "A market intelligence outage no longer tells you your own API key was rejected",+ "summary": "The two market intelligence reads in the unified API, recent narrative intel and momentum-ranked projects, pass a key we hold to the upstream provider; a caller needs no key at all. When that provider stopped accepting our key, both reads answered \"unauthorized\" with the provider's own wording, so an agent reading the response concluded its own credentials were bad and retried with different ones forever, when nothing on the caller's side was wrong. Both reads now say plainly that the service is temporarily unavailable because this deployment's upstream key was rejected, which is a status clients already know to retry rather than re-authenticate against. The older market intelligence endpoints always classified this correctly; the classification now lives in one place, so the two doors cannot drift apart again, and the error tables in the API reference list every code either one can return.",+ "link": "/docs/api-reference",+ "tags": [+ "fix",+ "docs"+ ]+ },{"date": "2026-08-15","type": "update",@@ -92,6 +103,17 @@"security"]},+ {+ "date": "2026-08-15",+ "type": "update",+ "title": "Asking for a voice that does not exist now tells you so, instead of quietly speaking in a different one",+ "summary": "The voice API accepts a provider and a voice id from the voice catalog. If the id was not one that provider actually has, nothing said so: the request was answered with audio rendered in that lane's default voice, and on the metered lanes it was billed, so a single typo in a voice id came back as a stranger's voice at full price with a success status. A well-formed id for a voice that had been removed took the opposite route and failed as a server error, which reads as our fault and tells a client to retry something that will never work. Both now return a plain rejection that names the id you sent and the catalog address to pick a real one from, and that check runs before anything is charged. Leaving the voice id out still gets you the lane default on purpose. Asking the catalog for a provider that does not exist used to return an empty list that was indistinguishable from a provider being down; it now says which providers exist. ElevenLabs voices, whose catalog grows every time you clone or add one, are still decided by ElevenLabs, but an id it rejects is reported as your request being wrong rather than ours, with any charge refunded first. Voice cloning also now asks who you are before reporting anything about how this server is configured.",+ "link": "/docs/api-reference",+ "tags": [+ "fix",+ "improvement"+ ]+ },{"date": "2026-08-15","type": "update",@@ -102,6 +124,16 @@"fix"]},+ {+ "date": "2026-08-15",+ "type": "update",+ "title": "Claim Your Wallet stopped telling real traders they had no track record",+ "summary": "Pasting a wallet into Claim Your Wallet could answer \"not indexed yet\" to a wallet with thousands of indexed pump.fun trades. The page only accepted a wallet the reputation scorer had already graded, and that scorer runs on a rollup that lags the trade indexer by design, so a wallet the platform had watched trade for days was still turned away. Any wallet with a real on-chain record now renders its full ledger and window aggregates immediately, with the reputation panel honestly reading unproven until the grade lands. Three more fixes shipped with it. Win and loss counts now cover closed positions only, so a trader still holding sixty fresh bags no longer reads as sixty losses, and a new realized profit-and-loss figure sits beside the existing net figure that charges an open bag's entry as if it were written off. Asking for devnet returned the wallet's mainnet trades; both halves of the read are now scoped to the network you asked for. And if the database is unreachable, the page now says the record is temporarily unavailable and offers a retry, instead of publishing an empty scorecard that the CDN would then serve to everyone for the next few minutes.",+ "link": "/claim-wallet",+ "tags": [+ "fix"+ ]+ },{"date": "2026-08-15","type": "update",@@ -124,6 +156,16 @@"fix"]},+ {+ "date": "2026-08-15",+ "type": "update",+ "title": "Looking up an on-chain agent by its CAIP reference works again",+ "summary": "The public resolver that turns an ERC-8004 agent reference into its owner, token URI and verified agent card could not be reached at all. An agent reference ends in a slash before its token id, and the documentation told callers to percent-encode the whole reference, which turns that slash into %2F. The API refuses any address segment holding an encoded separator, because that is the same trick used to escape a path and reach files that were never meant to be served, so every correctly-formed lookup was turned away as an unknown address before it ever reached the resolver. The verification badge you can embed on your own site built its request in exactly that shape, so it never showed a result either. The resolver now accepts the reference with its slash as a normal part of the address, the badge asks for it that way, and the documentation shows the form that works. The protection against encoded separators is untouched. New tests resolve the address through the same routing the live site uses, so this cannot quietly break again.",+ "link": "/docs/api-reference",+ "tags": [+ "fix"+ ]+ },{"date": "2026-08-15","type": "update",@@ -357,6 +399,16 @@"fix"]},+ {+ "date": "2026-08-15",+ "type": "update",+ "title": "The Tutor’s running tab and end-of-session invoice now show what you actually paid",+ "summary": "The Tutor charges one cent per answer and keeps an itemized tab you can reload later and close for a signed invoice. That ledger only ever knew how to store itself in a key-value service this deployment does not run, so nothing was ever written down: reopening the page never restored your history, and ending a session always produced an invoice reading \"no questions\" and a total of zero, however many answers you had paid for. The tab is now stored in the platform’s own database, so it survives a reload, a new browser tab, and any server restart, and the invoice lists every answer with its charge and a verification hash over the line items. Closing a session twice returns the identical invoice instead of a freshly stamped one, closing a session id that was never used no longer makes that id unusable for its real owner, and if the ledger itself is ever unreachable the page says so rather than quietly reporting that you owe nothing. Asking a question still succeeds even if the ledger write fails, because a bookkeeping outage should never cost you the answer you paid for.",+ "link": "/tutor",+ "tags": [+ "fix"+ ]+ },{"date": "2026-08-15","type": "update",diff --git a/public/changelog.xml b/public/changelog.xmlindex b876fdbfa..5c334d276 100644--- a/public/changelog.xml+++ b/public/changelog.xml@@ -14,6 +14,14 @@<category>fix</category><description>Payment sessions let you hand an agent a budget instead of a key, and two parts of that path judged an endpoint by one rule: if it did not ask for payment, the call was free. That is only true when the endpoint actually succeeded. A paid API answering 404, 400, or 500 was being scored as a free call, and the consequences ran in the wrong direction in both places. The dry run at /pay/simulator, whose whole job is to tell you in advance whether a budget and an allowlist will carry a workload, reported a plan as feasible at a cost of nothing when in fact every endpoint in it was erroring: free calls are never priced, never allowlist-checked, and never counted, so a wall of failures looked like a clean run. And an agent spending through a session got back a success envelope saying no payment was needed for an endpoint that had just failed on it. Both now separate the two cases: a real success with no payment challenge is still free and still costs nothing, while an error status is reported as an endpoint failure carrying the status the endpoint actually returned. No session is charged either way. The simulator lists each failing endpoint by name in its problems list and no longer calls the plan feasible.</description></item>+ <item>+ <title>A market intelligence outage no longer tells you your own API key was rejected</title>+ <link>https://three.ws/docs/api-reference</link>+ <guid isPermaLink="false">2026-08-15:A market intelligence outage no longer tells you your own API key was rejected</guid>+ <pubDate>Sat, 15 Aug 2026 12:00:00 GMT</pubDate>+ <category>fix</category>+ <description>The two market intelligence reads in the unified API, recent narrative intel and momentum-ranked projects, pass a key we hold to the upstream provider; a caller needs no key at all. When that provider stopped accepting our key, both reads answered "unauthorized" with the provider's own wording, so an agent reading the response concluded its own credentials were bad and retried with different ones forever, when nothing on the caller's side was wrong. Both reads now say plainly that the service is temporarily unavailable because this deployment's upstream key was rejected, which is a status clients already know to retry rather than re-authenticate against. The older market intelligence endpoints always classified this correctly; the classification now lives in one place, so the two doors cannot drift apart again, and the error tables in the API reference list every code either one can return.</description>+ </item><item><title>A retried agent spend can no longer hand its transaction to somebody else, and an unsupported chain says so instead of blaming the network</title><link>https://three.ws/docs/permissions</link>@@ -70,6 +78,14 @@<category>fix</category><description>The arm page clamps the risk knobs on the server so a bad number can never reach the trading loop, and those clamps have a floor: the smallest trade the loop will place is 0.001 SOL. That floor was being applied in the wrong direction. Ask to arm a live agent with a per-trade size of zero, and instead of being told no, you were armed for real spending at 0.001 SOL a trade with a half-SOL daily ceiling you never asked for. The check meant to catch this looked at the number after the clamp had already raised it, so it could never fire. Arming live now refuses any per-trade size below the floor, and any daily cap smaller than a single trade, and tells you which number to fix; nothing is armed until you do. Clamping a too-large number back down to the ceiling is unchanged, and simulate runs keep the same forgiving behaviour as before. The arm page also shows the reason a save was rejected instead of one generic line, so an ownership problem no longer reads the same as a bad trade size, and a failed Telegram test now prints what Telegram actually said.</description></item>+ <item>+ <title>Asking for a voice that does not exist now tells you so, instead of quietly speaking in a different one</title>+ <link>https://three.ws/docs/api-reference</link>+ <guid isPermaLink="false">2026-08-15:Asking for a voice that does not exist now tells you so, instead of quietly speaking in a different one</guid>+ <pubDate>Sat, 15 Aug 2026 12:00:00 GMT</pubDate>+ <category>fix</category>+ <description>The voice API accepts a provider and a voice id from the voice catalog. If the id was not one that provider actually has, nothing said so: the request was answered with audio rendered in that lane's default voice, and on the metered lanes it was billed, so a single typo in a voice id came back as a stranger's voice at full price with a success status. A well-formed id for a voice that had been removed took the opposite route and failed as a server error, which reads as our fault and tells a client to retry something that will never work. Both now return a plain rejection that names the id you sent and the catalog address to pick a real one from, and that check runs before anything is charged. Leaving the voice id out still gets you the lane default on purpose. Asking the catalog for a provider that does not exist used to return an empty list that was indistinguishable from a provider being down; it now says which providers exist. ElevenLabs voices, whose catalog grows every time you clone or add one, are still decided by ElevenLabs, but an id it rejects is reported as your request being wrong rather than ours, with any charge refunded first. Voice cloning also now asks who you are before reporting anything about how this server is configured.</description>+ </item><item><title>Avatars can move their lips again: the Audio2Face lane is back</title><link>https://three.ws/changelog</link>@@ -78,6 +94,14 @@<category>fix</category><description>Every call that turns speech into facial animation had been failing with an upstream not-found error, so a talking avatar spoke with a frozen face and the one-shot say() path returned nothing usable. The hosted model id the lane pointed at had stopped resolving for our NVIDIA account. It now points at the hosted Audio2Face NIM that is actually live, verified end to end: a synthesized clip drives a 30 fps ARKit blendshape track whose jaw genuinely opens and closes with the words. The discovery helper that finds these model ids was missing the name NVIDIA publishes them under, which is why the outage looked like a configuration mystery; it now finds them.</description></item>+ <item>+ <title>Claim Your Wallet stopped telling real traders they had no track record</title>+ <link>https://three.ws/claim-wallet</link>+ <guid isPermaLink="false">2026-08-15:Claim Your Wallet stopped telling real traders they had no track record</guid>+ <pubDate>Sat, 15 Aug 2026 12:00:00 GMT</pubDate>+ <category>fix</category>+ <description>Pasting a wallet into Claim Your Wallet could answer "not indexed yet" to a wallet with thousands of indexed pump.fun trades. The page only accepted a wallet the reputation scorer had already graded, and that scorer runs on a rollup that lags the trade indexer by design, so a wallet the platform had watched trade for days was still turned away. Any wallet with a real on-chain record now renders its full ledger and window aggregates immediately, with the reputation panel honestly reading unproven until the grade lands. Three more fixes shipped with it. Win and loss counts now cover closed positions only, so a trader still holding sixty fresh bags no longer reads as sixty losses, and a new realized profit-and-loss figure sits beside the existing net figure that charges an open bag's entry as if it were written off. Asking for devnet returned the wallet's mainnet trades; both halves of the read are now scoped to the network you asked for. And if the database is unreachable, the page now says the record is temporarily unavailable and offers a retry, instead of publishing an empty scorecard that the CDN would then serve to everyone for the next few minutes.</description>+ </item><item><title>Docs pages stop opening with a wall of raw metadata, and the AWS pages say where the Marketplace listing actually stands</title><link>https://three.ws/docs/aws-marketplace</link>@@ -94,6 +118,14 @@<category>docs</category><description>An audit of eight tutorials found instructions that could not work. The AI-brain tutorial described a model line-up that had been superseded: it never mentioned the Claude 5 family (Sonnet 5, Opus 5, Fable 5), which are live and selectable, and it recommended Sonnet 4.6 as the default. It also sent you to screens that do not exist, telling you to add API keys under a gear icon in My Agents and to change models, sampling knobs, and streaming under a Brain menu. Provider keys actually live in the AI Provider Keys panel on your account page, the model is a field on the agent's embed policy, and streaming has no off-switch. The promised usage tab with token charts and threshold emails did not exist either; the tutorial now documents the real per-agent usage read and points at the monthly quota and per-minute rate limit, which are the controls actually enforced before your money is spent. It also now explains that a visitor can never escalate your agent onto a paid model. The custom-GLB tutorial was worse for anyone following along: it linked a Studio address that no longer resolves, and four of its commands invoked gltf-transform subcommands that have never existed, so converting an FBX, renaming animation clips, and the full Mixamo walkthrough all failed at the first step. Those are replaced with the converter the platform itself uses and a clip-renaming script that was run against a real avatar to confirm it works. Its Draco flag was wrong, its upload path pointed at the wrong page, and it still warned that an avatar without an idle clip would stand frozen in a T-pose, which stopped being true when animation retargeting became universal. Finally, the page-events tutorial listed the wrong fallback order for emote animations.</description></item>+ <item>+ <title>Looking up an on-chain agent by its CAIP reference works again</title>+ <link>https://three.ws/docs/api-reference</link>+ <guid isPermaLink="false">2026-08-15:Looking up an on-chain agent by its CAIP reference works again</guid>+ <pubDate>Sat, 15 Aug 2026 12:00:00 GMT</pubDate>+ <category>fix</category>+ <description>The public resolver that turns an ERC-8004 agent reference into its owner, token URI and verified agent card could not be reached at all. An agent reference ends in a slash before its token id, and the documentation told callers to percent-encode the whole reference, which turns that slash into %2F. The API refuses any address segment holding an encoded separator, because that is the same trick used to escape a path and reach files that were never meant to be served, so every correctly-formed lookup was turned away as an unknown address before it ever reached the resolver. The verification badge you can embed on your own site built its request in exactly that shape, so it never showed a result either. The resolver now accepts the reference with its slash as a normal part of the address, the badge asks for it that way, and the documentation shows the form that works. The protection against encoded separators is untouched. New tests resolve the address through the same routing the live site uses, so this cannot quietly break again.</description>+ </item><item><title>Oracle stops answering "nothing is happening" when the database is the thing that is down</title><link>https://three.ws/oracle</link>@@ -262,6 +294,14 @@<category>docs</category><description>An audit of eight documentation pages ran every command they print and checked every claim against the code behind it, then corrected what had drifted. The biggest one was in the trading tutorial: the command it gave you for verifying a fleet's fills on-chain asked for a field the leaderboard does not publish, so it returned an empty explorer link for every agent and the verification step could not be completed at all. It now uses the fields that exist, linking both legs of each trade straight to Solscan, and it skips agents whose wallet has not been provisioned yet instead of building a URL that goes nowhere. The MCP tool safety page undercounted the tools that share a name across two servers, which matters because that collision is exactly why a trust policy has to key on the namespaced id; it now lists all nine and prints a command that re-derives the list so it cannot go stale again. The MCP tool catalog itself had drifted from its source and was serving outdated descriptions for three tools, on a page whose whole promise is that it is generated from source and cannot drift; it has been regenerated. The package extraction guide had every version in its roster out of date along with its build and test columns, and still described a publisher field as missing that was set some time ago. The 10 SOL trading experiment page stated a risk rule in absolute terms that contradicted the setup script it tells you to run, which read as a bug in one of the two; the rule now carries the scope the code actually applies.</description></item>+ <item>+ <title>The Tutor’s running tab and end-of-session invoice now show what you actually paid</title>+ <link>https://three.ws/tutor</link>+ <guid isPermaLink="false">2026-08-15:The Tutor’s running tab and end-of-session invoice now show what you actually paid</guid>+ <pubDate>Sat, 15 Aug 2026 12:00:00 GMT</pubDate>+ <category>fix</category>+ <description>The Tutor charges one cent per answer and keeps an itemized tab you can reload later and close for a signed invoice. That ledger only ever knew how to store itself in a key-value service this deployment does not run, so nothing was ever written down: reopening the page never restored your history, and ending a session always produced an invoice reading "no questions" and a total of zero, however many answers you had paid for. The tab is now stored in the platform’s own database, so it survives a reload, a new browser tab, and any server restart, and the invoice lists every answer with its charge and a verification hash over the line items. Closing a session twice returns the identical invoice instead of a freshly stamped one, closing a session id that was never used no longer makes that id unusable for its real owner, and if the ledger itself is ever unreachable the page says so rather than quietly reporting that you owe nothing. Asking a question still succeeds even if the ledger write fails, because a bookkeeping outage should never cost you the answer you paid for.</description>+ </item><item><title>The vanity receipt verifier stops calling honest receipts fraudulent, and the avatar CLI takes a Solana address</title><link>https://three.ws/docs/rig-doctor</link>diff --git a/public/features.json b/public/features.jsonindex 806558b00..22f5dba43 100644--- a/public/features.json+++ b/public/features.json@@ -1,5 +1,5 @@{- "generated_at": "2026-08-15T15:04:09.638Z",+ "generated_at": "2026-08-15T15:11:10.411Z","generated_by": "scripts/build-page-index.mjs from data/pages.json","site": {"name": "three.ws",