🍷 dsh-palate — an eye that grows

中文说明 · DeepSeek Harness plugin

Design-audit tools measure with a fixed ruler. dsh-palate trains an eye that grows.

Most design-review plugins ship a static ruleset and apply it forever — use them once or a thousand times, the judgment is identical. dsh-palate is the opposite: it keeps a taste corpus that accumulates. Every example you feed it and every principle you distill sharpens the judgment your agent draws on. The more you use it, the better its eye gets.

Why this exists

Taste is not a gift — it's pattern recognition built from exposure. See enough good and bad design, and the rules emerge. dsh-palate turns that into a mechanism an agent can actually use:

1. Feed — record designs you judged good or bad, and why
2. Distill — recurring lessons become codified principles
3. Review — critique a new design against the accumulated taste, not a generic checklist
4. Grow — every verdict reinforces the principles it supports; the palate compounds

What the agent gets

Tool	What it does
palate_review	Assemble the accumulated taste (principles + relevant past examples) as context, so the agent critiques grounded in learned judgment
palate_add	Feed an example (good/bad/note + reason + tags) into the corpus — grows the palate
palate_learn	Distill a new principle from experience and add it to the codified taste
palate_list	Browse the accumulated corpus
palate_principles	List the codified principles, ordered by evidence
palate_stats	How much taste has accumulated: examples studied, principles distilled

Ships with a starter palate of 12 foundational principles (hierarchy, contrast, type scale, spacing rhythm, alignment, palette discipline, affordance, feedback, clarity, and an anti-AI-slop rule), so it's useful out of the box — then it grows from there.

How it works

palate_add (good/bad + why)  ──▶  taste corpus (SQLite + Markdown mirror)
palate_learn (new rule)       ──▶  codified principles
palate_review (a design)      ──▶  principles + relevant examples  ──▶  agent writes grounded critique
        ▲                                                                │
        └──────────────────── verdicts reinforce principles ◀───────────┘

• Storage: node:sqlite (built into Node ≥ 22) at $DSH_HOME/palate/, plus human-readable taste.md / principles.md mirrors. Zero runtime dependencies.
• The panel: a draggable overlay shows the growth story — examples studied, principles distilled, recent judgments.
• Vision pairing: feed it screenshots by reading them with a vision tool first (e.g. modlens_read_image), then pass the description to palate_review.

Honest framing

This is accumulated retrieval + codified principles, not model fine-tuning. The plugin supplies the learned taste as context; the model renders the actual critique. That's deliberate: the judgment stays auditable (you can read taste.md and principles.md), and it improves as the corpus grows — without retraining anything.

Install

dsh plugin --profile web add github:guo6x/dsh-palate

Requirements: DeepSeek Harness web profile, Node ≥ 22. Restart dsh web, refresh the page — a 👁️ button appears at the sidebar foot.

Develop

pnpm install
node build.mjs        # esbuild → lib/index.js (host ESM) + lib/client.js (ModuleLoader bundle)
node tests/smoke.mjs  # 17 pure-logic checks (no browser needed)

MIT licensed. Ideas and examples welcome — open an issue.

Known limitations

• No semantic matching in the plugin itself — relevance between a subject and past examples is tag/keyword based; the model does the deeper reasoning from the assembled context.
• Markdown mirrors are read-only exports for v0.1 (human edit-and-merge-back is planned).
• Vision is delegated — pair with a vision tool to review screenshots.