Skip to content

Chore/skills catalog updates#8

Merged
deve1993 merged 4 commits into
mainfrom
chore/skills-catalog-updates
Jul 20, 2026
Merged

Chore/skills catalog updates#8
deve1993 merged 4 commits into
mainfrom
chore/skills-catalog-updates

Conversation

@deve1993

Copy link
Copy Markdown
Collaborator

No description provided.

OpenCode Bot and others added 4 commits July 20, 2026 14:50
…92 palettes, 22 stacks)

Vendors the latest ui-ux-pro-max-skill bundle from
github.com/nextlevelbuilder/ui-ux-pro-max-skill (MIT license) into all
three bundle locations: skills/ui-ux-pro-max/, packages/codegraph/data/
skill/ui-ux-pro-max/, and .agents/skills/ui-ux-pro-max/. data/, scripts/,
and references/ are wholesale-replaced to match upstream exactly (adds
google-fonts.csv, motion.csv, references/, validate_data.py, tests/, and
6 new stack files; renames web-interface.csv -> app-interface.csv);
local evals/ and learnings.md are left untouched. Also syncs the
lifecycle-skills SKILL.md-only mirror and adds a version: 2.11.0
frontmatter line to every copied SKILL.md.

MIT attribution: LICENSE + NOTICE (crediting nextlevelbuilder/
ui-ux-pro-max-skill) added to each of the 3 bundle locations.

Security review (scripts/*.py) before vendoring: no network calls, no
exec/eval/subprocess/os.system, no secret/credential/env access beyond
a benign COLORTERM check, and all file writes are confined to
<output-dir>/design-system/<slug>/ with path-traversal protection via
safe_slug(). Verdict: safe to vendor.

Local dev DB (.codegraph/graph.db, gitignored) updated surgically via
direct UPDATE of the ui-ux-pro-max row's content/description/updated_at
-- no full re-import. Prod skill routing (memory.fl1.it) is untouched
and needs the separate deploy pipeline in
packages/codegraph/docs/DEPLOY-SKILLS.md.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
…n skills, MIT)

Integrates 8 pure-knowledge SKILL.md files from Leonxlnx/taste-skill
(github.com/Leonxlnx/taste-skill, MIT License, 65k+ stars) into the
SkillBrain catalog, mirroring the frontend-design skill's 4-location
footprint (./skills, ./.agents/skills, ./packages/codegraph/data/skill,
./packages/codegraph/data/lifecycle-skills) plus a local dev-DB row each.

Skills added:
- taste-skill (v2.0.0) — anti-slop brief-inference skill for landing
  pages, portfolios, redesigns
- minimalist-skill — warm monochrome editorial minimalism
- brutalist-skill — industrial/tactical-telemetry brutalist UI
- soft-skill — high-end agency-grade visual polish
- redesign-skill — audit-first upgrade of existing sites
- image-to-code-skill — image-first design-to-implementation workflow
- output-skill — full-output enforcement, bans truncation/placeholders
- brandkit — premium brand-kit/identity image generation

Upstream frontmatter `name:` values were normalized to match each
directory slug (e.g. design-taste-frontend -> taste-skill); descriptions
kept verbatim for FTS routing. A `version:` line was added per skill
(2.0.0 for taste-skill per upstream README, 1.0.0 for the rest, no
upstream version markers found). Each file carries an MIT attribution
footer; the upstream LICENSE is stored once at
packages/codegraph/data/skill/taste-skill/LICENSE.

Content was scanned for prompt-injection / exfiltration patterns
(ignore-instructions phrasing, curl|sh, credential/env reads, hidden
HTML-comment directives) before inclusion — none found; all 8 files are
pure design-guidance prose.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
…llmark v1.1.0)

Single SKILL.md (build/audit/redesign/study modes, 57 slop-test gates).
Vendored into the 4 catalog mirrors + local DB. MIT attribution in footer + LICENSE.
skill-guard dogfood: SAFE (score 20); lone P1 match is hallmark's own anti-injection
guidance line ('override system/developer/user instructions'), a benign false positive.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
… lanes

Rewrote the description frontmatter (body untouched) of 5 overlapping
design skills across all 4 mirror copies so skill_route separates them
into distinct lanes:

- frontend-design: component-level craft/polish toolkit (20 refinement
  commands) for EXISTING UI — not for greenfield pages or data-backed tokens.
- taste-skill: aesthetic-direction umbrella (vibe/style inference) for
  handcrafted-feeling landing pages/portfolios.
- hallmark: anti-AI-slop skill with explicit build/audit/redesign/study
  modes — owns the audit/redesign/study routing keywords.
- ui-ux-pro-max: data-backed design-system generator (styles, palettes,
  fonts, charts, stacks) — owns design system/palette/tokens routing.
- redesign-skill: taste-skill-family premium redesign of existing sites,
  complementing hallmark's redesign mode.

Each new description also cross-references sibling skills so an agent
landing on the wrong one is pointed to the right lane. DB row (content,
description, updated_at) synced to match on the local graph.db.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
@deve1993
deve1993 merged commit 0a2fa14 into main Jul 20, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant