18d5f5bfc0720ba0a5b42a630644b4d6cbe0143e
16 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 claude
|
18d5f5bfc0 |
Make Piggy part of the product rather than a guest in it
Piggy arrived as a chat panel bolted onto a CRM and then grew a workspace around it. The layout was already right — the audit found the approval card to be the best-designed object in the repo, and the account page's empty panels less finished than anything in the workspace. What was wrong was vocabulary: nobody had written the small things down, so both halves kept inventing them. Piggy was drawn with five different marks — a pig in the dock, a sparkle in the sidebar and again on the model picker, a speech bubble on the Ask buttons, and a stock robot glyph on every assistant message, which is the one people look at most. There is now one mark. The composer, which is the first control in the product since sign-in lands on /piggy, was the only un-adapted shadcn field left: 6px radius against a 12px Send button it sat 8px from. A stat tile had been reinvented six times at three numeral scales, and the same uppercase micro-label existed in five variants, two of them one tab apart in the same rail. There were 63 hand-written font sizes: not a scale, sixty-three opinions. Underneath that, the focus ring was invisible. The global rule used ring-accent, which Tailwind deliberately aliases onto the hover tint, so the ring measured 1.01:1 against the light canvas — no visible focus indicator anywhere in the product, for any accent, in either theme. It is ring-brand now and measures 17:1. The warning, positive and info tones were darkened until each clears 4.5:1 on a card, on inset and on its own chip, and the light canvas moved to 98% so a card lifts without leaning on its shadow. The mobile work is the part worth reading. A landscape phone gave the transcript 28% of the viewport and a keyboard-up phone 16%, against a 45% floor — and the fixed tab bar painted over the composer, covering the safety sentence and half the Send button, because two source comments asserted the bar stood down on short viewports and it never had. Both fixed and measured by hit-testing rather than by screenshot. The composer itself was 64px tall for a blank second line nobody typed, because the auto-resize effect sizes to scrollHeight and scrollHeight counts rows — a CSS height could not win against an inline style, so the attribute was the honest lever. Verified across both themes driven through the app's own control: no horizontal overflow on 15 routes at four viewports, 672 stat values that fit, 297 labels at exactly 11px/500, Escape returning focus to its opener rather than the body on every overlay, and a rejected write no longer reporting "Succeeded" with a green check. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|
claude
|
f0173440e4 |
Put Piggy on Prime Agent, and let it write to the book
Piggy was a hand-rolled OpenAI tool loop. It is now a Prime Agent session — Prime Intellect's own harness, embedded as a Node library — answering from PIG's tools and, for the first time, able to put information into the CRM rather than only read it out. The harness is a coding agent, so the first job was taking the coding agent away from it. `noTools: 'all'` plus an explicit allowlist leaves the model with PIG's ten `pig_*` tools and no bash, no filesystem, no IPython. That holds under attack: a hostile extension, a skill and a settings file planted in the agent's own directory, then `setActiveToolsByName` called with every built-in, still leaves ten tools, all ours. Both lines are load-bearing — `noTools` alone registers nothing, and the allowlist is what admits our own. Writing is gated rather than assumed. A change is proposed, not made: the tool returns a description, the transcript renders a diff card, and nothing reaches the database until someone presses Apply. Contracts, commitments, allocations and compliance always stop for a human whatever the mode. Every write runs through `executeMutation` as the calling user, so their capabilities and the audit trail apply exactly as they would to a human's. Four things about the SDK are wrong in its own documentation and cost a debugging cycle each: models.json does not resolve an env var name for `apiKey`, it sends the literal string; there is no built-in prime-inference provider in 0.84.1; a ResourceLoader you pass in is never reloaded for you; and the stock system prompt is a coding-assistant prompt that must be replaced — but replacing it also silently removes the tool list, because the harness only renders that section when it owns the prompt. AGENTS.md records all four. The expensive one was thinking level. The harness defaults to `medium`, and nemotron spent an entire 4,096-token budget reasoning and returned an empty answer. `low` was worse; `off` omits the parameter so the endpoint's default wins. An explicit `reasoning_effort: none` via `thinkingLevelMap` took a turn from 6,195 output tokens to 149. And a turn is now bounded. The harness loop is `while (true)` with no iteration cap; a runaway on a frontier model would have eaten the credit it is supposed to report on. Ceilings on model calls and tokens, enforced both through the harness hook and independently from the event stream, plus a per-user daily spend limit — and the ledger now records spend on turns that fail, which it previously discarded. Signing in lands on /piggy, which is a workspace: conversations down one side, the agent in the middle, what it did and what it cost beside it. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|
claude
|
99d165b5e5 |
Rebuild Piggy's interface, and give the demo book a business to describe
Piggy answered in raw markdown, threw away every tool result it streamed, and fought the reader's scroll on every token. The three surfaces that made it worth having — what it read, how it reasoned, what it cost — were all on the wire and none of them reached the screen. The transcript is now composed of five parts under components/piggy: answers render through streamdown, the container sticks to the bottom without pinning the reader there, tool steps say what they read and link to the record, and each turn carries its model and token count. Three lifecycle bugs went with them: Stop left a permanent spinner, a truncated stream was indistinguishable from thinking, and a failed send destroyed the message it failed to send. Underneath, the inference path grew timeouts, jittered retries on 429 and 5xx, tolerance of the malformed frames a 30B model emits, and an agent_runs row per turn so chat spend is observable. The system prompt now states that a field ending in Cents is cents — without it nemotron renders costPerGpuHourCents: 189 as "$189 per GPU-hour", which is a 100x error on the most scrutinised number in the room. The demo book was arithmetically incoherent: every deal's value contradicted its own allocation revenue by up to 3.6x, nothing had ever closed, no customer had any paper, and the marketplace was empty. Deal value is now derived from the allocation, the book clears 5.3% across five blocks with one deliberately underwater, and the renewal, compliance and agent-provenance machinery finally has rows to act on. A --clear that deleted every obligation, SLA term and capacity request in the database regardless of origin is scoped to the demo's own ids. Around that: accounts have a detail page, ⌘K searches the book, Settings can mint the API keys it always claimed to, and deploy.sh actually ships the agent instead of silently skipping its compose profile. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
 karti
|
9db53cb36f |
Polish the public auth experience
Replace the stacked sign-in card with a responsive public shell, a local monochrome compute field, restrained grain and reduced-motion-safe drift. Share the anonymous header with Learn so public navigation stays consistent, and carry the corrected spacing through registration and profile setup. |
||
|
karti
|
afce0dda28 |
Play the platform music on the shared Learn page too
Reversing yesterday's call at Karti's direction. The provider moves from Shell up above the router in App, so a share-code visitor gets the same character as a member rather than a silent page. The reason that is safe is the same reason "autoplay" was never really autoplay: the browser refuses audio until the page has had a real gesture, so nothing plays the instant a link opens — it starts once someone is actually using the page. The anonymous page renders OUTSIDE Shell and therefore has no app header, so the mute control is added to its own chrome. Music with no way to stop it is the worst version of this feature, and a visitor who cannot find the switch does not conclude the site has taste. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|
karti
|
0108a70131 |
Add platform music: three tracks, looped, mutable
Three ~28s tracks Karti generated, re-encoded from 160kbps to 96kbps and pulled from -15 LUFS to -26. They were mastered at foreground level; the failure mode of background music in a tool someone has open for eight hours is not "too quiet". Playback volume is a further 0.14 on top. AUTOPLAY DOES NOT MEAN AUTOPLAY. Every current browser refuses audio with sound until the page has had a real user gesture — Chrome sometimes relents for a site with a high Media Engagement Index, Safari essentially never does on a first visit. So `play()` rejects on mount, and the naive version of this looks like a bug: the control says playing and nothing is audible. This tries immediately, and on refusal arms a one-shot listener and starts on the first click or keypress. Verified in Chrome: paused on load, playing 2.2s after the first click. That also happens to be the kind behaviour for someone who opened six tabs at once. Mounted in Shell, NOT in App. The anonymous Learn page renders outside Shell, and a share-code visitor opening a link someone sent them should not get unexpected audio — that is the one context where it reads as a fault rather than as character. Asserted by there being no <audio> element on that page at all. Preference is localStorage and deliberately not mirrored to the server, on the same reasoning as the sidebar: whether you want music depends on whether you are wearing headphones, not on who you are. Also pauses on tab hide, because music from a tab nobody is looking at is the thing people hunt through twenty tabs to kill. The element is rendered rather than `new Audio()` so it is inspectable in devtools and in a test; `preload="none"` means a user who mutes it never downloads a track. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|
karti
|
19dd30acbe |
Give the Learn cards a real frame instead of a gradient
The preview cards led with a generated gradient. It was a deliberate fallback — nothing renders a frame of a Cap embed without loading the embed, and loading nine embeds to decorate a grid is how a page becomes unusable on a phone — but for videos PIG serves itself the frame is right there in the file. The poster is named after the VIDEO's content hash, not its own: `overview.4d4581ae.mp4` -> `overview.4d4581ae.jpg`. Re-rendering a clip changes both names together, so a thumbnail cannot outlive what it claims to show. It needs no schema column and no manifest entry, because the name is derivable. `learnPoster.sh` cuts the frame with `thumbnail=90` starting four seconds in rather than taking frame 0: the first frame of a Playwright capture is often mid-paint, and a poster of a half-rendered page is worse than no poster. The resolver ASSERTS the poster rather than verifying it — @pig/core is pure and has no filesystem. That is safe in both directions: a missing poster 404s, which `<video poster>` renders exactly as it renders no poster, and which the card falls back from via onError. Claiming a poster that is absent is free; omitting one that exists would cost every card its thumbnail. Cap-hosted rows are unchanged and still get the gradient, verified by there being exactly five <img> elements on a page with nine resources. Also widens the media allowlist to jpg/webp. The filename pattern, the traversal rules and the symlink check are untouched and still cover them, because extension is the only axis that changed. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|
karti
|
45b70b17f0 |
Redesign Learn, and give it five real videos in Karti's voice
THE PAGE. The anonymous route rendered outside Shell, so it sat flush against
the viewport edge and read as a form rather than a product — which is the first
thing anyone at Prime Intellect sees when the link is shared. It now brings its
own chrome and leads with a hero; the platform track is a numbered course, the
concept tracks are a poster grid, and admin add/archive moved behind one Manage
toggle so they stop competing with the content. Verified in Chrome at 1440 and
393, light and dark: horizontal overflow is 0 in all three access states.
THE VIDEOS. Five ~30s walkthroughs, narrated in Karti's cloned voice through
Chatterbox and cut against real screen capture of the seeded demo book. The
audio is rendered FIRST and its measured duration drives the capture, because a
shot list that runs short leaves the narrator talking over a frozen frame and
one that runs long gets cut mid-sentence. Levels are loudness-normalised so
clips do not jump between videos.
Cap cannot take a programmatic upload — video.karti.ai needs an interactive
login — so PIG serves these itself. A native <video> on this origin needs no
iframe and therefore no CSP frame-src at all; Karti's own Cap recordings still
render through the existing iframe path, which is why the resolver is now a
discriminated union.
THREE THINGS THE VERIFIERS CAUGHT, all of which shipped green:
- createMediaRoutes was never mounted. Every layer landed — migration, seed,
both feeds, the bind mount, the docs — except the one that serves the bytes,
so /media/learn/* fell through to the SPA fallback and answered HTTP 200
text/html. The player showed a black box with working controls and no error.
The tests certified the route factory in isolation, which proves the handler
and says nothing about whether it is wired in. There is now an assertion
against the ASSEMBLED app, and it fails loudly on content-type — the failure
mode is a 200, not a 404.
- A symlink in the media directory escaped the root. resolve() is lexical and
stat() follows links, so the containment check this file's own header
promised did not hold. realpath before the check closes it.
- Vite proxied only /api, so self-hosted playback broke for anyone running the
app the documented way — in the same invisible 200-text/html manner.
Also: a duplicate media slug used to throw from the middle of seedDemo() and
take out every later section; it now reports and skips that one entry. And the
player has an onError state, because content-addressed filenames mean a
re-render deliberately leaves the old row pointing at a file that is gone.
The three DEMO platform rows are dropped — five real recordings supersede them,
and placeholders sitting under real ones made the page read as half-finished to
the audience it is meant to convince. The supply and demand concept rows stay:
there are no real recordings for those tracks yet, and an empty track hides the
shape of the page.
Tests 275, typecheck clean.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
||
|
karti
|
13dec6b4b8 |
Rebuild the shell, add Calendar and Learn, and govern reads
Seven parallel agents and an adversarial verification pass. The three things worth knowing before reading the diff: RBAC WAS ALREADY BUILT. docs/build-plan.md marks F2 and F3 outstanding and is stale — packages/core/src/permissions.ts and lib/mutation.ts shipped long ago. So this does not rebuild them; it closes the gaps an audit found. The big one is that reads were entirely ungoverned: every GET was "any authenticated member", so a junior demand rep and a research contractor could both pull per-block supplier cost and break-even prices from /api/capacity/margin, and every contract's negotiated terms. For a company whose margin is the business, that was the hole that mattered. Adds book:read / economics:read / team:read, a readGuard middleware, and a `viewer` role below member. THE BUTTON AND THE 403 DISAGREED — the exact thing F3 said must never happen. Contracts.tsx never called can() at all, so its save button was always enabled against a server requiring contract:sign; Capacity.tsx gated commitment creation on deal:write/demand while the server wanted commitment:write/supply. POST /api/activities was the one write bypassing executeMutation: no capability check, and any member could mutate accounts.lastActivityAt as a side effect. It is now a proper mutation() behind activity:write. The shell becomes three panes — a collapsible shadcn sidebar with an account switcher on the Piggy accent, a header with real search, and Piggy docked to the right, page-aware and persistent across navigation. The phone keeps its bottom tab bar, which is the thing this product already beat trycompai/crm on, and gains the sidebar as a sheet. Calendar is a projection over thirteen dated sources rather than a new table, because a table would duplicate dates that already live on contracts, deals and commitments and would drift — and one ledger answering the question is the whole argument. It surfaces export_authorizations and compliance_artifacts, which had indexed expires_at columns, schema comments saying they must be alerted on, and no read endpoint or UI anywhere. Learn carries two tracks. Concepts are members-only; the platform track can be opened with a share code by someone with no account. The code mints a scoped learn-only token and never a Principal — every route here resolves a principal and then checks capabilities, so a principal-minting code would be one missing check away from leaking the book. "Only platform-track rows may be code-visible" is a database CHECK constraint as well as a write-path rule, and a test asserts a valid learn token still gets 401 on /api/dashboard, /api/accounts and /api/contracts — the same invariant scripts/deploy.sh refuses to ship without. CD becomes tag-to-ship. CI publishes an image to the Gitea registry on a release-* tag and cloud-2 pulls it, so no credential on the shared runner can execute anything on production — by construction rather than by policy. Both halves of deploy.sh's original rule survive: nothing on the runner reaches the host, and a human still decides when it ships. deploy.sh gains a rollback and a public-origin check, and PIG_IMAGE now reaches compose through `sudo env`, without which sudo's env_reset silently resolved every release to pig:local. Tests 141 -> 261. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|
karti
|
e12d27edd1 |
Polish every product workflow across desktop and mobile
CI / verify (push) Successful in 3m32s
Reframe each screen around the decisions compute brokers make: sellable capacity, full-cost margin, pipeline movement, contract deadlines, evidence review, staged imports, and controlled agent access. Group the shell by operating domain, strengthen mobile navigation and sheets, add responsive record treatments, and make loading, error, empty, readiness, and retry states explicit. The visual audit exposed sortable table targets and an unnamed file input only after exercising the rendered app, so this commit also pins those accessibility decisions at their actual interaction boundaries. Manrope is self-hosted as a single Latin variable subset to keep the stronger hierarchy without shipping unused font payloads. |
||
|
karti
|
1318c0b841 |
Ship growth intelligence and demo polish
CI / verify (push) Successful in 3m51s
|
||
|
karti
|
c2c7fb9c19 |
Make mutations confirm themselves, and seed the evidence trail
CI / verify (push) Successful in 2m51s
Two demo gaps, both of which made working features look like they were not there. **Toasts fired into nothing.** RecordSheets already called toast.success on every save, but <Toaster /> was never mounted, so nothing appeared. It could not be mounted, either: the shadcn original imports next-themes, which PIG does not use — it has its own provider so a chosen theme is persisted server-side and follows a user between devices. Rewired to PIG's useTheme, mounted inside ThemeProvider, and offset clear of the phone tab bar and the home indicator. Feedback added where the interface otherwise gives none: allocation and hold report the GPU-hours actually written, because the sheet closes on success and the only other evidence is a number moving off-screen; releasing a hold says the capacity is sellable again; fact decisions say what the decision meant, and that approving evidence is not the same as writing it to a record; the profile form confirms rather than just clearing itself, which otherwise reads as the input being discarded. **The fact table was empty**, so the review queue and every provenance tooltip had nothing to show — the mechanism that makes an agent-written CRM trustworthy, invisible. Six agent-derived facts seeded with a deliberate mix: two applied, showing what a confident agent writes unprompted, and four proposed, including one weak claim that a reviewer should reject, so the queue is not a row of obvious approvals. Each carries a score, a band, evidence and where available a source. Idempotent on subject+field+value; verified over two runs. Verified: toast confirmed firing in a real browser on a 393px viewport, 135 unit tests and e2e green, typecheck clean, CSP hash unchanged, 0px horizontal overflow across 12 routes at both breakpoints. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|
karti
|
2763531ce4 |
Align shadcn's accent token with what shadcn means by it
CI / verify (push) Successful in 2m53s
shadcn uses `bg-accent` for its SUBTLE surfaces — dropdown item hover, command row selection, ghost and outline button hover, the dialog close affordance. The brand colour in shadcn is `primary`. PIG's Tailwind config mapped `accent` to `--accent`, which is the brand. That inverted the meaning, so every shadcn hover and selection state painted a full-strength brand block. With the monochrome "pig" palette in dark mode the brand is near-white, so a selected command row rendered as a white slab against a near-black sheet. Measured before the change: selected row rgb(250,250,250) on a rgb(9,9,11) body. `accent` now aliases `--accent-subtle` and `accent-foreground` aliases `--accent-fg`, which is what those tokens were created for. The eleven places where PIG's own components wanted a solid brand fill — filled chips, selected card borders, progress bars — move to `primary`, which still resolves to `--accent`. A `brand` alias is added for clarity. After: selected row rgb(39,39,42) in dark and rgb(244,244,245) in light, both a subtle tint above the body; the pipeline's active stage chip stays a solid rgb(250,250,250) fill, unchanged. Found by opening overlays, which earlier screenshot sweeps never did — every route had been checked, but a dropdown or a command palette only misbehaves once it is open. Worth remembering: page-level sweeps do not exercise portals. Typecheck clean, 135 unit tests and e2e green, CSP hash unchanged, 0px horizontal overflow across 12 routes at 393px and 1440px. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|
karti
|
6bd5526675 |
Give Card min-w-0 so the page stops scrolling sideways on a phone
The Overview page overflowed 80px at 393px wide. Traced to the "The book" card: the grid column was a correct 361px, the card inside it was 457px and refused to shrink. Confirmed by forcing `min-width: 0` on grid children in the live page, which took the overflow to 0. Fixed on the Card base class rather than at the call site, because this is the third time the same trap has been fixed individually — grid and flex children default to `min-width: auto` and cards routinely hold something unshrinkable, a tabular-nums figure or a nowrap badge. `min-width: 0` is inert for a block-level card outside a flex or grid parent, so applying it always costs nothing and removes the whole class of bug. Verified by running the stack locally against the demo data: 0px overflow across all 12 routes at both 393px and 1440px. AGENTS.md updated to say any NEW container primitive needs the same, with the one-line browser check to confirm it. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|
karti
|
853bde2265 |
Build the agent-native compute CRM platform
CI / verify (push) Successful in 3m6s
|
||
|
karti
|
de33a03524 |
Add the web app, seed data, and user-selectable theming
apps/web — React, Vite, Tailwind, shadcn-idiom components. Mobile Safari is a first-class target, not an afterthought: - Two navigation treatments rather than one compromise. A bottom tab bar on phones, because the top of a large phone is out of thumb reach; a persistent sidebar from lg upward, so an iPad in portrait gets it too. - Safe-area insets throughout, so the tab bar clears the home indicator and the last row of a list is actually reachable. - Inputs are pinned to a 16px minimum, which is the correct fix for Safari zooming on focus. user-scalable=no is not used: it breaks pinch-zoom for everyone and recent iOS ignores it anyway. - The pipeline board becomes a stage picker on phones. An eight-column board scrolling horizontally on a 390px screen is technically responsive and practically useless. Theming: users pick an accent and the whole interface re-tints. Accent values live once, in @pig/core, and are written onto the root element at runtime — there is no CSS copy to drift from the TypeScript. Preferences are stored server-side so they follow a person between laptop and phone, mirrored into localStorage only so the pre-paint script can avoid a white flash. Status colours stay fixed regardless of accent: if "at risk" re-tinted to whatever someone picked, the signal would be gone. Seed data is public research, every record carrying a confidence grade and a source URL. No email addresses are seeded or inferred — none are published, and guessing them from a name and a domain is unreliable and rude. Authorship is not promoted to employment: contributors, residency participants and alumni are recorded as what the evidence actually shows, and a name that could not be sourced at all is listed as unresolved rather than invented. Three defects found and fixed by actually running it rather than assuming: 1. The seed was not idempotent. onConflictDoNothing() with no target is a no-op without a matching unique constraint, so a second run duplicated 27 contacts. There is deliberately no unique index on (account, name) — two people at one company can share a name — so idempotency is enforced in the seed instead of by bending the schema. 2. /capacity scrolled sideways on a phone. Grid items default to min-width:auto and `truncate` sets nowrap, so a long title became unshrinkable content and widened the track. Fixed with min-w-0 on every truncating grid child. 3. The idle-capacity alert silently failed to fire at exactly 80% utilisation, losing a float comparison against a 0.2 threshold. Moved to 0.15, which is also a more sensible line for "worth attention". The worked example is tuned to teach rather than to flatter: 70% sold at a 53% markup lands at +6.7% margin with 20% still idle, so both the healthy number and the alert are visible. Drop the sold share to 55% and the same block goes underwater — that sensitivity is the argument for the product. Verified in a real browser at 393px and 1440px, light and dark: zero horizontal overflow on every route, zero console errors. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> |