56bf6ee1b6
* feat: pre-generation research (Tavily) for grounded generation
Adds an optional pre-generation research step so the agent can produce
slides / prototypes / decks grounded in real sources instead of guessing.
User flow:
1. Settings -> Tavily Search -> paste API key (or set TAVILY_API_KEY).
2. Click the new Research button in the chat composer.
3. On send, the daemon runs a Tavily search, prepends the findings
as a <research_context> block ahead of the system prompt, and
spawns the agent. Research progress shows up as status pills in
the chat stream; the agent cites sources inline as [1]/[2]/...
Phase 1 surface:
- Single provider (Tavily), single depth ('shallow'), no LLM
synthesis pass (Tavily's `answer` is the summary).
- Composer toggle only; no popover / depth picker yet.
- Reuses the existing `status` SSE agent payload + StatusPill UI
so no new event variants or renderer code are needed.
Layers touched:
- contracts: ResearchOptions / Source / Findings DTOs;
ChatRequest.research; export from index.
- daemon: apps/daemon/src/research/{index,tavily}.ts orchestrator
+ provider; tavily added to MEDIA_PROVIDERS and ENV_KEYS; hook
in startChatRun before prompt assembly.
- web: ChatComposer toggle + ChatSendMeta; threaded through
ChatPane / ProjectView / streamViaDaemon into ChatRequest.
Side fix (required to land the feature, but useful on its own):
contracts internal relative imports lacked the `.js` suffix that
NodeNext module resolution requires. This was already breaking
`pnpm --filter @open-design/daemon typecheck` on main; without the
fix, none of the new research types were visible to the daemon.
All internal contracts imports now carry `.js`.
Spec: specs/current/research-feature.md (phases 2-4 outlined for
follow-up: composer popover, multi-provider, deep recursion, example
skills with research_recommends).
Verified:
- pnpm --filter @open-design/contracts typecheck/test
- pnpm --filter @open-design/daemon typecheck (the chokidar
project-watchers test is a pre-existing flake, unrelated)
- pnpm --filter @open-design/web typecheck
- node scripts/verify-media-models.mjs
* fix(daemon): clamp Tavily max_results to 20
Tavily's /search endpoint requires `max_results` in [0, 20]; sending a
larger value (e.g. when `research.depth: "deep"` resolves to 30) returns
400 and `runResearch` silently falls back to no-research. Clamp at the
provider boundary so Phase 2 depth tiers above 20 still produce results
instead of failing the request.
Generated-By: looper 0.6.1 (runner=fixer, agent=claude-code)
* Remove stale research merge leftovers
* Add agent-callable research search
* Fix Indonesian locale typecheck
* Fix research command invocation edge cases
* Harden slash search prompt expansion
* Honor research source caps in command contract
* Require search reports in design files
* Add research data provider settings
* Wire web research provider fallback order
* Update research provider fallback wording
* Revert "Update research provider fallback wording"
This reverts commit
|
||
|---|---|---|
| .. | ||
| accessibility-baseline.md | ||
| animation-discipline.md | ||
| anti-ai-slop.md | ||
| color.md | ||
| form-validation.md | ||
| laws-of-ux.md | ||
| README.md | ||
| rtl-and-bidi.md | ||
| state-coverage.md | ||
| typography.md | ||
Craft references
Brand-agnostic craft knowledge. Each file is a small, dense rulebook on one dimension of professional UI craft (typography, color, motion, …). Skills opt into the references they need; the daemon injects only the requested ones into the system prompt above the active skill body.
Why a third axis next to skills/ and design-systems/
| Axis | Scope | Example |
|---|---|---|
skills/ |
Artifact shape | saas-landing, dashboard, pricing-page |
design-systems/ |
Brand visual language (the 9-section DESIGN.md) |
linear-app, apple, notion |
craft/ |
Universal craft knowledge — true regardless of brand | letter-spacing rules, accent-overuse caps, anti-AI-slop |
DESIGN.md tells the agent which colors and fonts a brand uses. craft/
tells the agent the universal rules a competent designer applies on top —
e.g. ALL CAPS always needs ≥0.06em tracking, regardless of the brand.
How a skill opts in
Add an od.craft.requires array to the skill's front-matter. Only the
listed sections are injected, so a skill that needs only typography pays
no token cost for color/motion content.
od:
craft:
requires: [typography, color, anti-ai-slop]
Allowed values match the file names in this directory minus the .md
extension. Unknown values are silently ignored (forward-compatible).
Why silent fallback instead of fail-fast?
A skeptical reader will ask: "If a skill requests a planned-but-not-yet-vendored
section and the corresponding file doesn't exist yet, shouldn't we warn
the user?" We chose forward-compatibility over fail-fast: a skill
authored today can list a planned slug and start benefiting the moment
the matching craft/<slug>.md is vendored in a follow-up PR, with no
skill edit needed. The cost of a missed reference is a missing
paragraph in the system prompt, not a broken skill — so the loud
failure mode is not worth the friction.
Note for skill authors arriving from older guidance: an earlier draft
used motion as the future-slug placeholder. The shipped equivalent
today is animation-discipline. Use that one if your skill emits
motion.
Enforcement levels
Craft files mix auto-checked rules and guidance.
- Auto-checked. Rules wired into
apps/daemon/src/lint-artifact.ts— currently the P0 list inanti-ai-slop.md(Tailwind-indigo accent, two-stop hero gradients, emoji-as-icons, etc.). The linter reports these as findings back to the UI (for P0/P1 badges) and to the agent (as a system reminder for self-correction). Artifact persistence is not currently hard-blocked on P0 hits. - Guidance. The rest. The agent reads the rules, reviewers apply them, the linter doesn't check them.
A purely behavioral craft file (state-coverage, animation-discipline) is guidance unless a specific rule is later promoted into lint-artifact.ts.
Files
| File | Section name | When to require |
|---|---|---|
typography.md |
typography |
Any skill that emits typed content (~all skills) |
color.md |
color |
Any skill that emits styled output (~all skills) |
anti-ai-slop.md |
anti-ai-slop |
Marketing pages, landing pages, decks |
state-coverage.md |
state-coverage |
Any skill with stateful UI (dashboards, mobile apps, forms, list/table views) |
animation-discipline.md |
animation-discipline |
Any skill that ships motion: mobile apps, multi-screen flows, gamified UI, transitions, microinteractions |
accessibility-baseline.md |
accessibility-baseline |
Any skill that ships interactive UI: dashboards, forms, mobile flows, anything with focus/labels/keyboard paths |
rtl-and-bidi.md |
rtl-and-bidi |
Any skill that ships localized text or layout: blogs, docs, financial tables, mobile apps, anything that may render Arabic / Hebrew / Persian |
form-validation.md |
form-validation |
Any skill whose primary artifact contains an interactive form: lead capture, sign-in, signup, settings, multi-step intake |
laws-of-ux.md |
laws-of-ux |
Any skill whose composition decisions hit named cognitive limits: pricing pages (Hick's, Choice Overload, Von Restorff), dashboards (Pareto, Selective Attention, Working Memory), onboarding (Goal-Gradient, Zeigarnik, Peak-End), modals (Fitts's, Tesler's). Sibling axis to the rendering-rule files above — covers what to compose, not how to render. |
Partial-stateful skills. A skill that's mostly static but contains an embedded form, data table, or query surface should opt in. State-coverage rules apply to the stateful component, not the whole page.
More sections (icons, craft-details) will be added in follow-up
PRs as we wire the linter side.
Attribution
Craft content is adapted from the MIT-licensed
refero_skill project
(© Refero Design), with edits to fit Open Design's house style and link
back to OD's design tokens (var(--accent) etc.) instead of generic
Tailwind hex values.