opencode-skills-collection 4.1.10 → 4.1.12

package/bundled-skills/reverse-engineer-anything/references/evidence-workflows.md

@@ -0,0 +1,40 @@
+# Evidence, comparison, and verification workflows
+
+Evidence-producing tools return the complete Evidence record directly in MCP
+`structuredContent` and CLI JSON. Read `normalized_result` for operation data,
+and cite `evidence_id` when another tool or your explanation needs a stable
+reference. Preserve authority, limitations, coverage, and residual unknowns. If a result
+is incomplete or paginated, continue only when the remaining data matters to
+the task; do not fetch a bundle or resource merely to read a result already
+returned inline. When complete `inspect_binary_layout` or
+`analyze_javascript_application` Evidence is already retained, call
+`inspect_analysis_view` for a summary, one section/symbol/module, or a stable
+page instead of repeating the producer.
+
+Use `record_unknown` to track unresolved questions in the user's investigation
+and name the authority or environment still required. Supply supporting and
+contradicting evidence IDs. It updates the session registry directly.
+Use `update_unknown` with the current revision; reread after a stale revision
+instead of retrying blindly. Only qualifying observed evidence can verify a
+resolution.
+
+Use comparisons with complete, compatible page sets when claiming equivalence
+or absence:
+
+- `compare_artifacts` compares inventory pages by occurrence path, content,
+  metadata, and graph relations.
+- `compare_functions` compares explicit function Evidence, not fuzzy
+  whole-binary matches.
+- `compare_bundles` compares canonical bundle membership and unknown history.
+- `find_changed_behavior` combines existing comparisons; static differences
+  remain candidates, not causal proof.
+- `build_call_path` needs explicit function dossiers from one artifact/provider;
+  an incomplete frontier makes absence unknown.
+- `correlate_static_and_runtime` uses explicit mappings; matching patterns are
+  hypotheses, not causality.
+- `verify_reconstruction` evaluates a finite typed specification. A pass covers
+  only declared comparable claims, not global equivalence.
+
+Process captures are opt-in behavioral evidence, not a security sandbox. V3
+captures cannot be upgraded to V4; rerun the original scenario. Distinguish root
+exit from descendant settlement and require freshness when the task needs it.

package/bundled-skills/reverse-engineer-anything/references/javascript-applications.md

@@ -0,0 +1,70 @@
+# JavaScript and Electron application artifacts
+
+Use `analyze_javascript_application` directly on the operator-supplied ASAR or
+extracted tree. The complete result, graph, and Evidence context are returned
+inline when they fit the MCP frame. Source-map contents are part of the static analysis. Start from its
+findings, coverage, unknowns, and graph context, then make a focused follow-up
+only when a specific question remains unanswered.
+
+An oversized response reports `resource_constraint` and
+`details.resource: "transport"`. Reuse
+`details.reported_limits.evidence_reference` with `inspect_analysis_view` for a
+summary, one module, or a stable page of module identities. Use
+`trace_application_feature` once a module, route, or string seed is known.
+`export_evidence_bundle` writes the complete canonical session to a
+caller-selected file. These workflows keep the original Evidence and coverage;
+a broad follow-up can also exceed framing.
+
+BrowserWindow preferences, preload and contextBridge surfaces, IPC
+registrations, utility processes, and native binding requests are static syntax
+observations. Only a unique exact literal IPC channel match supports an inferred
+pairing. Dynamic or ambiguous channels remain unresolved. A requested `.node`
+member is not a verified native export. Never claim runtime reachability,
+registration, defaults, or policy enforcement from static analysis.
+
+Use `trace_application_feature` on existing application Evidence for one
+literal node ID, route, string, API, IPC channel, module, or native export.
+Choose a direction and reuse the returned application Evidence ID when the
+connected server advertises retained references, or supply complete Evidence
+inline. Include
+Hopper or Ghidra Evidence only when its artifact digest matches exactly.
+
+For version comparison, analyze each version once, then call
+`compare_application_versions`. Accept only its digest, source-map, structural
+fingerprint, or non-module semantic matches. Module ordinals and minified names
+are not persistent identity. Report added or removed only with complete
+opposite-side coverage; otherwise report unknown.
+
+When the question asks how one exact exported callable's returned object shape
+changed, analyze each version once and then call
+`compare_javascript_export_shapes` with explicit module paths and export names.
+Use the returned IDs on the same connection, or complete inline Evidence
+records, from both analysis calls. Accept variant
+pairing only through the tool's unique exact literal discriminant. Read
+`property_inventories` and each change's `presence` before treating `unknown` as
+"the name was not observed": inventories list observed property names even when
+values stay unknown, with a `source_range` for each paired or unpaired variant.
+Array holes and mutation-invalidated slots are not observed properties.
+Complete parent coverage reports presence-only
+`added`/`removed` independently of unresolved values. Cite the
+comparison Evidence and report JSON Pointer changes; dynamic values, ambiguous
+variants, and incomplete parent-property coverage stay unknown. This is static
+inference, not runtime behavior. When runtime semantics are needed, run
+behavioral probes against the relevant application versions and capture them
+through the available browser, Electron, or process workflows.
+
+## Reusing application Evidence
+
+When advertised by the connected server, application trace and compare tools
+accept complete inline Evidence or
+`{"kind":"retained-evidence","evidence_id":"ev_<64 lowercase hex characters>"}`
+for their application input (`application`, or `left`/`right`). `inspect_analysis_view`
+uses the same retained-reference form in `source`, or portable inline Evidence,
+to project a summary, module page, or one module with its recorded observations. This notation is
+a template: replace it with the actual returned ID. Versions before 4.1.0 accept only full
+inline Evidence. Use the exact ID
+returned by the producer on the same MCP connection. Resolution does not run
+analysis or select a provider; findings remain inline. Native observation
+arrays still take full Evidence. `close_binary` clears retained references;
+export a bundle before closing, or supply portable inline Evidence in another
+connection. A missing reference includes its ID and recovery guidance.

package/bundled-skills/reverse-engineer-anything/references/native-and-artifacts.md

@@ -0,0 +1,146 @@
+# Native, managed, and packaged artifacts
+
+## Recorded Linux crashes
+
+Use `inspect_recorded_crash` with an explicit Linux x86-64 ELF core path to
+inspect historical thread registers, signals, raw notes and original file
+ranges. It is independent of the active disassembler target. Source-bound
+register values and note bytes are observations; missing notes, signal-thread
+association and current executable/library identity remain unknown. PIDs are
+historical metadata and never authorize live attach or control.
+
+The optional `include_debugger_context` facet adds mapping candidates through
+caller-supplied GDB/pwndbg in an owned core-only session. Display names do not
+establish current file identity; zero reported flags leave permissions unknown.
+Requested unavailable context fails with setup guidance. See
+[recorded crashes](https://github.com/morluto/rea/blob/main/docs/recorded-crashes.md)
+for exact upstream profiles, bounds and verification coverage.
+
+## Offline ELF layout
+
+Use `inspect_binary_layout` with an explicit local ELF path for file-backed
+sections, symbols, relocations, linkage names, and static mitigation inferences
+without opening a disassembler target. When that complete record is already
+retained, call `inspect_analysis_view` for a summary, mitigations or linkage
+facet, one section or symbol, or a stable page. Do not re-run the decoder to
+read a single object.
+
+## Native targets
+
+After `open_binary`, use focused search, procedure, or function tools directly.
+Use `binary_overview` when target metadata or inventory context helps answer the
+question. Prefer literal search, names, decompilation, callers, callees, and
+cross-references. Addresses and recovered pseudocode are analysis observations,
+not original source. Provider unavailability and unsupported metadata remain
+unknown rather than false.
+
+To see which functions or Objective-C methods a Mach-O actually calls in one
+run, use `observe_native_calls` with explicit breakpoints and a bounded
+`duration_ms`/`max_events`. It launches a new process under LLDB, so the target
+runs and may change files or use the network. Hardened-runtime targets need the
+`get-task-allow` entitlement.
+
+Use `binary_session` with no arguments to check the open target, selected
+provider, and alignment. Its default `result.tool_availability` includes the
+complete tool inventory with availability reasons and remediation. When choosing
+a tool, use its entry in that result; `tools/list` retains the complete catalog
+when the target or provider state changes.
+
+## DOS MZ with Ghidra
+
+For an admitted DOS MZ target, use `inspect_native_load_image` to check measured
+loaded bytes, file mappings, relocations and the entry against the immutable
+snapshot. Keep `mismatch` and `unsupported` explicit. Verified import covers the
+reported static image, not runtime DOS or PC-98 hardware behavior.
+
+Use `read_bytes` for initialized analysis memory and `address_to_file_offset` to
+anchor one observed address to original file bytes. Loader fixups can change a
+word; a source offset does not imply byte equality. Complete function body ranges
+use inclusive ends and can contain gaps. Do not treat the enclosing span as code.
+For packed targets, retain the original and separately derived artifact identities;
+decompiling an unpacking stub does not recover the unpacked program.
+
+When reading Objective-C dispatch metadata, distinguish class implementation
+methods from protocol requirements: class methods report null requirement
+flags, while decoded protocol declarations retain required and optional facts.
+Property attributes remain raw name/value tokens; use the property's
+`atomicity` field for the decoded `N` marker or the default atomic behavior.
+Null class/meta flags, ivar counts, atomicity, or read-only values mean the
+selected provider did not establish that fact. Symbol names alone do not prove
+that a class is a root class or has no ivars.
+
+## Managed PE/CLI
+
+Start with `inspect_managed_artifact`. REA's canonical managed inspection is
+execution-free: do not claim it loaded, reflected, executed, or resolved the
+assembly. Keep managed/native boundaries and unavailable reconstruction facts
+explicit. A bring-your-own reconstruction oracle is separate from the canonical
+parser and must not become an implicit setup dependency.
+
+### .NET NativeAOT
+
+NativeAOT binaries contain native machine code, so ordinary CIL decompilers
+cannot recover the original method bodies. When managed metadata/CIL is absent,
+classify that as a native-analysis route rather than claiming source recovery.
+Open the native image with the selected deep provider and inspect functions,
+strings, calls, and references; describe decompiler output as pseudocode and
+inference.
+
+NativeAOT images can be PE/COFF, ELF, or Mach-O executables and shared
+libraries. Use file bytes and loader metadata rather than suffixes to identify
+them. For packaged apps, inventory/extract with REA's artifact tools and retain
+the exact contained image identity. PDB, ELF debug, or dSYM sidecars are
+additional evidence only when the provider accepts and matches them.
+
+For deeper type recovery, the optional third-party
+[Ghidra NativeAOT analyzer](https://github.com/Washi1337/ghidra-nativeaot)
+rehydrates ReadyToRun metadata and can recover method tables, type relationships,
+frozen objects, and strings. This is a Ghidra extension with its own install and
+interactive workflow; it is not currently part of REA's managed parser or
+headless Ghidra bridge. Treat its annotations as analysis evidence, retain the
+binary identity, and do not present recovered types as original source. Header
+discovery may require a symbol or analyst identification of the ReadyToRun
+header, depending on the binary and analyzer version.
+
+## Packages and extraction
+
+Use `inspect_artifact` for application bundles, archives, ZIP/APK/IPA/MSIX/AppX,
+ASAR, or DMG inputs when the artifact graph and findings help answer the
+question. It returns the complete artifact graph inline. Cite graph manifest
+IDs when using them.
+
+For an IPA or a macOS `.app`, ZIP, or DMG, pass the inventory Evidence to
+`project_apple_application_graph`. It reports application roots and nested
+bundles by path convention: app extensions, XPC services, frameworks, login
+items, system and driver extensions, and plug-ins. It also lists privileged
+helpers, launchd plists, symlinks, and each bundle's `info_plist_path` and
+executable candidates. Roles are path conventions, not parsed plists; read the
+listed plists with `inspect_plist`.
+
+After `open_binary` on a `.app` or Mach-O, use `trace_dylib_resolution` to see
+which file each `@rpath`, `@loader_path` and `@executable_path` load reaches for
+every executable in the bundle. It also lists missing or weak loads and earlier
+`@rpath` candidates that are absent. Narrow large bundles with `roots` or
+`architecture`. System paths stay undetermined because the dyld shared cache
+provides them.
+
+Use `extract_artifact` when materialized files are needed. It takes no arguments
+and materializes all regular files into a fresh temporary directory chosen by
+REA. Symlinks and encrypted entries are inventory facts, not extractable files.
+The result reports the materialized directory; callers do not choose the
+destination. No separate permission grant is needed.
+
+Native DMG traversal automatically uses a read-only mount on supported macOS
+hosts, with an owned temporary mount directory and cleanup. It needs no approval
+flag. Unsupported hosts or mount failures remain explicit limitations; do not
+claim child inventory when only root identity is available. Extraction is a
+separate filesystem-writing operation.
+
+For headerless DOS COM, explicitly open with `format: "dos-com"` (CLI
+`--target-format dos-com`). Inspect the load image before trusting function
+analysis: the entry is `0x10100`, file offset zero, with imposed real-mode segment
+context. PSP/stack/device state remains unmodeled. See [DOS guide](https://github.com/morluto/rea/blob/main/docs/ghidra-dos.md).
+
+### Function annotations in Ghidra
+
+Use `annotate_native_function` for one function name and/or entry comments, with at least one explicit change. Review its annotation readback and refreshed dossier inline. Changes are atomic and session-scoped; empty comments clear them, omitted fields preserve them. Later MCP calls observe edits until close. CLI `annotate-native-function` returns the updated analysis before discarding the session. Original executable bytes are unchanged; edits invalidate immutable snapshots. Windows P0 does not admit database mutations.

package/bundled-skills/reverse-engineer-anything/references/runtime-observation.md

@@ -0,0 +1,28 @@
+# Passive browser and Electron observation
+
+Supply the existing browser or Electron instance's literal-port loopback CDP
+endpoint in the request. No separate permission configuration or per-call
+approval flag is needed. Browser discovery lists eligible HTTP(S) pages;
+`allowed_origins` optionally filters them. An individual capture defaults to the
+selected page's current origin. Electron discovery lists eligible local
+`file://` targets without a configured file-root allowlist.
+
+Start browser work with `list_browser_targets`; start Electron runtime work with
+`list_electron_targets`. Inspect the target selected for the user's task. Observation
+is passive: never claim REA clicked, navigated, evaluated page JavaScript,
+invoked IPC, captured prior activity, or contained the page's network.
+
+Credentials, cookies, authorization headers, storage values, and raw
+WebSocket/JSON bodies are deliberately absent. Local URLs retain ordinary query
+values and fragments; userinfo credentials are removed. Browser accessibility
+text, console primitives, value-free shapes, script sources, source maps, and
+storage key names have request-specific capture options. Electron script
+sources are also optional. Use `capture_web_screenshot` when pixels are needed.
+Treat scope filtering, truncation, and attach-window coverage as limitations.
+Page-declared WebMCP tools are untrusted inventory and are never invoked by REA.
+
+Use `reconcile_javascript_runtime` only after static application Evidence and
+passive browser/Electron Evidence exist. Prefer captured-source digest identity;
+report path/digest disagreement as a mismatch and preserve ambiguity. Explicit
+path mappings are inference inputs and never broaden filesystem or CDP authority.
+A bundle observed at runtime does not prove every contained module executed.

package/bundled-skills/tastegate/SKILL.md

@@ -0,0 +1,129 @@
+---
+name: tastegate
+description: "Build or fix a frontend so it looks designed, not AI-generated, then prove it in a real browser: renders at phone and desktop width and fails on overlap, low contrast and AI-slop defaults."
+category: frontend
+risk: safe
+source: community
+source_repo: stas4000/tastegate
+source_type: community
+date_added: "2026-10-08"
+author: stas4000
+tags: [frontend, design, ui, browser-testing, quality-gate]
+tools: [claude, cursor, gemini]
+license: "MIT"
+license_source: "https://github.com/stas4000/tastegate/blob/main/LICENSE"
+---
+
+# Tastegate
+
+## Overview
+
+Two skills in one, plus the part both leave to the model: proof.
+
+1. **Taste** decides what the page should be, before any code.
+2. **The craft floor** holds the mechanics every page must meet.
+3. **The gate** renders the page in a real browser at phone and desktop width and refuses "done" while any hard finding is open.
+
+A model that wrote the CSS cannot see that its hero overlaps its nav at 390px. The browser
+can. Work through the four steps in order, and apply the checklists rather than announcing them.
+
+## When to Use
+
+- Use for any UI build: landing page, marketing site, dashboard, app screen, component
+- Use for a redesign, a polish pass, or a "make it look better" request
+- Use when a page needs to stop looking AI-generated
+- Use before calling any frontend work done, as the proof step
+
+## How It Works
+
+### Step 1: Read the room (2 minutes, before code)
+
+Write one line before building:
+
+> Reading this as: `<page kind>` for `<audience>`, in a `<vibe>` language, built on `<type pairing>` + `<palette family>` + `<one signature detail>`.
+
+- The audience picks the look, not your habit. A B2B buyer, a design-savvy consumer and a public-sector visitor need three different pages.
+- Set three dials from that read (1 to 10): **variance** (symmetry to art), **motion** (still to cinematic), **density** (gallery to cockpit). Default landing page: 7 / 6 / 4. Trust-first or regulated: 3 / 2 / 5.
+- Name one **signature detail** the page will be remembered by: a type treatment, a material, an interaction, a way the product is shown. One, done well.
+- If the brief pins fonts, colors or an era, the brief wins over every rule here.
+
+### Step 2: Build on the craft floor
+
+- **Type:** a real display face with character, paired with a readable text face. Not Inter, Roboto, Arial or the system stack as the display voice unless the brief asks for neutral. Body 16px or more, 60 to 75 characters a line, clear size and weight steps.
+- **Color:** one neutral family, one accent, locked across the page. No purple-to-blue "AI glow" gradients, no gradient text, no colored glow halos. Secondary text tinted from its surface, never plain gray on color. Body contrast 4.5:1 or better.
+- **Layout:** the hero fits the first screen with its call to action visible. No small uppercase eyebrow label over every heading. No row of three same-size cards as the page's structure. Each section uses a different layout family. Spacing tighter inside a group than between groups.
+- **Icons and copy:** real SVG icons in one stroke weight, never emoji. The product's own words: no lorem ipsum, no "Acme", no "John Doe", no "Unlock the power of".
+- **States:** hover, focus-visible, active, disabled, loading, empty and error exist for every control that has them. Style the browser parts too: selection color, focus ring, scrollbar, caret.
+- **Motion:** one authored moment, eased out, from an already-visible default. Respect `prefers-reduced-motion`.
+- **Phone first:** every multi-column block declares what it becomes under 768px. Tap targets at least 44px. Nothing scrolls sideways.
+
+### Step 3: Run the gate
+
+```bash
+python3 skills/tastegate/scripts/gate.py <path/to/index.html or http://localhost:3000> --out .tastegate
+```
+
+(Run it from the repo root, or use the path to wherever this skill is installed.)
+
+It renders the page in headless Chromium at 390x844 and 1440x900, prints every finding with
+its rule, viewport and element, saves full-page screenshots, and exits 1 while any **fail**
+finding is open. First run only: `pip install playwright && python3 -m playwright install chromium`.
+
+Fix every fail finding in one batch, then run the gate again. Warnings are judgment calls:
+fix them unless the brief asked for that thing.
+
+### Step 4: Look at it yourself
+
+Open both screenshots the gate saved (`.tastegate/desktop.png`, `.tastegate/mobile.png`). The
+gate catches mechanics; it cannot tell you the page is boring. Ask: would a designer at a good
+studio ship this? Does the signature detail land in the first viewport? Is anything still a
+default you did not choose? Fix what you see, rerun the gate, then stop.
+
+## Examples
+
+### Example 1: A landing page that passes review but fails the gate
+
+The copy and sections are right, so the work reads as done. The gate at 390px reports the
+hero heading overlapping the nav, a 3.1:1 contrast ratio on secondary text, and sideways
+scroll from an un-wrapped feature row. Three real defects, none visible in the markup.
+
+### Example 2: "Make it look less AI"
+
+The gate names what "AI slop" actually is on this page: a purple-to-blue gradient hero,
+gradient heading text, emoji feature icons, an uppercase eyebrow over every section, the
+system font stack as the display voice, and three equal cards carrying the structure. Each
+is a line item to fix, not a matter of opinion.
+
+## Best Practices
+
+- ✅ Write the direction line before the first line of CSS
+- ✅ Fix all fail findings in one batch, then rerun the gate
+- ✅ Open both screenshots yourself after the last change
+- ✅ Report the gate's last line and the screenshot paths with the result
+- ❌ Don't call a page done on a passing build or a glance at the markup
+- ❌ Don't let a default font, palette or card grid decide the page
+
+## Limitations
+
+- The gate checks mechanics. It cannot tell you the page is boring, off-brand or wrong for the audience: step 4 is not optional.
+- It needs Playwright and headless Chromium installed, and a page it can reach (a file path or a local server).
+- A brief that pins fonts, colors or an era overrides the craft floor.
+- Does not replace accessibility auditing by a human or real user testing.
+
+## Security & Safety Notes
+
+- `scripts/gate.py` launches headless Chromium and loads the URL or file you give it. Point it at your own page or a local server, in a local or authorized test environment; it will fetch whatever that page fetches.
+- It is read-only against your project: it renders, measures and writes screenshots plus findings into the `--out` directory. It does not edit source files.
+
+## Files
+
+- `scripts/gate.py` - the browser gate (needs `playwright` and headless Chromium)
+- `references/direction.md` - the full step 1 direction rules
+- `references/craft-floor.md` - the full step 2 craft floor, read it right before the first UI edit
+
+## Credits
+
+Adapted from [stas4000/tastegate](https://github.com/stas4000/tastegate) (MIT). The direction
+rules derive from taste-skill (MIT) and the craft floor from impeccable (Apache-2.0); both
+notices ship here in [references/THIRD_PARTY_NOTICES.md](references/THIRD_PARTY_NOTICES.md) with their
+full licenses under `references/`.

package/bundled-skills/tastegate/references/THIRD_PARTY_NOTICES.md

@@ -0,0 +1,25 @@
+# Third-party notices
+
+tastegate's own code (scripts/gate.py, tests) and SKILL.md are MIT, Copyright (c) 2026 Bles Software.
+Two reference files are adapted from other open-source skills. Both were rewritten, shortened and
+merged for tastegate; the originals are much larger and worth reading in full.
+
+## impeccable
+
+- Source: https://github.com/pbakaus/impeccable (commit 9d715cc4f5564a990ca8345abfdd5df6dc9b41c8)
+- Author: Paul Bakaus
+- License: Apache License 2.0, full text in impeccable-LICENSE-Apache-2.0.txt
+- Used in: references/craft-floor.md, adapted from the "Craft floor" reference (verify list and
+  refuse list). Changes: rewritten in our words, merged with taste-skill's layout rules, marked the
+  rules gate.py checks, removed references to impeccable's CLI and hooks.
+
+## taste-skill
+
+- Source: https://github.com/Leonxlnx/taste-skill (commit ce26fc25c0e5e8cab638f883de62d9a86ee5e45b)
+- Author: Leonxlnx
+- License: MIT, full text in taste-skill-LICENSE-MIT.txt
+- Used in: references/direction.md (brief reading, the three dials, type and palette direction,
+  layout families) and parts of references/craft-floor.md (hero, eyebrow, card-grid and CTA
+  rules). Changes: rewritten and cut down to about a tenth of the original.
+
+tastegate is not affiliated with or endorsed by either project.

package/bundled-skills/tastegate/references/craft-floor.md

@@ -0,0 +1,53 @@
+# Craft floor
+
+Adapted from the craft floor in impeccable by Paul Bakaus (Apache-2.0, see THIRD_PARTY_NOTICES.md) and the layout rules in taste-skill (MIT), rewritten and merged. Load it right before the first UI edit. The brief's own words can earn back anything under "Refuse"; your habit cannot.
+
+Rules marked **[gate]** are checked by `scripts/gate.py` on the rendered page. The rest are yours to check in the screenshots.
+
+## Verify on the built page
+
+- **Contrast [gate]:** body and placeholder text 4.5:1 or better, large text (24px, or 18.7px bold) 3:1. Secondary text on a colored surface is tinted from that surface, not gray.
+- **No sideways scroll [gate]:** at 390px nothing is wider than the screen. Long words, code, tables and images get a wrap or a scroll container.
+- **No text over text [gate]:** no heading under a sticky nav, no badge on a label, no line of copy running into another at any width.
+- **Tap targets [gate]:** buttons and controls at least 44px tall on phones; inline links at least 24px.
+- **Readable size [gate]:** no visible text under 12px on phones.
+- **Hero [gate]:** the H1 and the first call to action are both inside the first desktop screen.
+- **Images [gate]:** every image loads and has alt text (empty alt for decoration).
+- **Type:** body measure 60-75 characters, display no bigger than about 6rem, tracking no tighter than -0.04em, balanced headings (`text-wrap: balance`), obvious steps in size and weight.
+- **Spacing:** tight inside a group, generous between groups, more space above a heading than below it.
+- **Depth:** shadows have an offset and a soft blur and take their tint from the background. A zero-offset colored glow is decoration.
+- **Motion:** one authored moment, eased out, from a visible default state. Scattered fades on every section are not motion design. `prefers-reduced-motion` turns it off.
+- **States:** hover, focus-visible, active, disabled, loading, empty, error. Forms: label above input, error below, no placeholder as label.
+- **Browser surfaces:** text selection, focus ring, caret, scrollbar, link underline offset, tabular numerals in data. Theme them; they are the cheapest sign a page was built rather than assembled.
+- **Copy:** the product's own words. Buttons say what they do. One label per intent on the whole page ("Get started" everywhere, not also "Try free" and "Sign up").
+- **Nav:** one line on desktop, 64-80px tall.
+- **Coverage:** every requirement in the brief is on the page and easy to find.
+
+## Refuse (the defaults that make a page look generated)
+
+Page scaffolds:
+
+- **[gate] Three or more same-size cards of icon, heading and text** as the page structure. Cards are for real elevation; group with space and dividers otherwise. Never nest cards.
+- **[gate] An eyebrow label** (small uppercase letter-spaced text) above headings. At most one per three sections, and usually none: the heading can speak.
+- The hero-metric template: giant number, small label, three supporting stats.
+- Section numbers (01 / 02 / 03) unless the order carries meaning.
+- A split header (big heading left, small paragraph right) with nothing else in the right column.
+- A logo wall or tagline stuffed inside the hero.
+- A modal for a task that does not need to interrupt.
+
+Surface habits:
+
+- **[gate] Purple, violet or indigo gradients** as the default accent (the "AI glow").
+- **[gate] Gradient text.** Emphasis comes from weight, size or the accent color.
+- **[gate] Emoji or unicode glyphs as icons.** Use one SVG icon set in one stroke weight.
+- **[gate] The system or a stock font as the display voice** (Inter, Roboto, Arial, Helvetica, system-ui) unless the brief asked for neutral.
+- **[gate] Placeholder copy:** lorem ipsum, Acme, John or Jane Doe, example.com, "Unlock the power of".
+- **[gate] Em dashes in body copy.** Use a comma, a colon or a period.
+- Glass and blur as decoration rather than a specific effect.
+- A thick colored border on one side of cards, callouts or list items.
+- Hard offset block shadows outside a page that is actually neo-brutalist.
+- Monospace as a costume for "technical" rather than for code or data.
+- Geometric masks faking an organic cut-out of a photo.
+- Mixed corner radii with no rule. Pick sharp, soft or pill and keep it.
+
+The floor holds the mechanics. It never picks the direction. With every check green, spend the page on the signature detail you chose in step 1.

package/bundled-skills/tastegate/references/direction.md

@@ -0,0 +1,59 @@
+# Direction: decide what the page is before building it
+
+Adapted from taste-skill by Leonxlnx (MIT, see THIRD_PARTY_NOTICES.md), rewritten and cut down.
+
+Most generated UI looks the same because the model skips this step and reaches for its default: a centered hero on a dark mesh, a purple-to-blue gradient, Inter on slate, three equal feature cards, an eyebrow label on every section. Every one of those is fine when chosen. None of them is fine as a reflex.
+
+## 1. Read the signals
+
+1. **Page kind:** SaaS landing, consumer product, agency, event, portfolio, docs, dashboard, settings, redesign.
+2. **Vibe words the user used:** minimal, calm, editorial, Linear-like, Apple-like, playful, brutalist, premium, serious B2B, dark tech.
+3. **References:** URLs, screenshots, named competitors. These beat your taste.
+4. **Audience:** a procurement panel, a design-conscious consumer, a developer, a recruiter skimming, a patient. The audience picks the look.
+5. **Assets that already exist:** logo, colors, type, photos. In a redesign they are starting material.
+6. **Quiet constraints:** accessibility-first, public sector, regulated, kids, trust-first commerce. These override aesthetics.
+
+Then write the one-line read from SKILL.md step 1. If two very different reads are plausible, ask one question. If not, proceed.
+
+## 2. Set the dials
+
+| Read | Variance | Motion | Density |
+|---|---|---|---|
+| Minimal, calm, editorial | 5-6 | 3-4 | 2-3 |
+| Premium consumer | 7-8 | 5-7 | 3-4 |
+| Playful, experimental, agency | 9-10 | 8-10 | 3-4 |
+| Landing page (default) | 7 | 6 | 4 |
+| Trust-first, regulated, public | 3-4 | 2-3 | 4-5 |
+| Dashboard, admin, tool | 3-4 | 2-3 | 6-8 |
+| Redesign that preserves | match | +1 | match |
+
+- **Variance above 4:** avoid the centered hero. Use a split, left copy with a right asset, asymmetric space, or a pinned scroll section. A centered hero is right for a manifesto or launch note where the words are the design.
+- **Motion:** 1-3 means hover and focus only. 4-6 means one entrance moment and scroll-linked reveals. 7+ means a real authored sequence, still with a reduced-motion path.
+- **Density above 7:** no generic cards; numbers breathe in plain layout with dividers.
+
+## 3. Pick the type
+
+- Choose the display face for its character, then a text face that reads well at 16px.
+- Sans display by default: Geist, Satoshi, Cabinet Grotesk, General Sans, PP Neue Montreal, Switzer, Space Grotesk, Manrope, Outfit. Rotate; do not reuse the same pairing on consecutive projects.
+- A serif needs a reason you can say in one sentence (editorial, heritage, luxury publication). "It feels premium" is not a reason. The two LLM-favorite display serifs (Fraunces, Instrument Serif) are not defaults.
+- Emphasis inside a headline is italic or bold of the same family, not a random second family.
+- Italic display words with descenders (g, j, p, q, y) need line-height 1.1 or more so they do not clip.
+
+## 4. Pick the palette
+
+- One neutral family (zinc, stone, slate, sand, ink), one accent, saturation under 80%. Lock the accent for the whole page: no blue CTA in a warm page's footer.
+- Families that are not the default: pure monochrome with one bright pop, cobalt and cream, forest with an amber accent, terracotta and slate, olive and brick, black and tan, cold silver and smoke.
+- The premium-consumer default (cream background, brass or clay accent, espresso text) is a reflex. Use it only when the brand is really that.
+- Light or dark comes from the use scene: who, where, under what light. Not from the category.
+
+## 5. Pick the layout families
+
+List the sections, then give each a different family: split, full-bleed media, bento with real visual variety, marquee, stacked statement, comparison table, timeline, quote, gallery, sticky stack, horizontal pan. The same family at most once. Never three image-and-text zigzags in a row.
+
+## 6. Pick the signature detail
+
+One thing a visitor will remember: an oversized numeral treatment, a material (paper grain, frosted glass used on purpose, machined metal), a product demo that actually moves, a cursor interaction, a custom illustration style. Spend the effort there. Everything else stays quiet and precise.
+
+## Redesigns
+
+First decide: evolve or replace. Evolve keeps the identity, copy and behavior and fixes the mechanics. Replace treats the old look as an anti-reference and picks a new world with the steps above. Never split the difference by polishing a look you have decided to throw away. Never change factual copy, prices or claims without asking.