CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

visual-edit

Open and collaboratively edit a running local app in Design, with shared fallback previews and source handoff. Use when the user asks to inspect, share, or edit a real local app in Design.

Quality

83%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

The canonical home for this skill is visual-edit in BuilderIO/skills

SKILL.md
Quality
Evals
Security

Quality

Content

77%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

An exceptionally actionable and well-validated skill body with copy-paste commands, payloads, and strong feedback loops for every risky operation. Its weaknesses are length and structure: duplicated warnings inflate the token budget, and host-specific, security-model, and writeback details that belong in one-level-deep reference files are all inlined with no bundle files at all.

Suggestions

Split the body into reference files (e.g. references/browser-webmcp.md for per-host browser instructions, references/react-writeback.md for provenance and anchor rules, references/security-model.md for the account/sharing capability model) and keep SKILL.md as a concise overview with clearly signaled one-level-deep links.

De-duplicate repeated guidance: the frame-count limits in "Useful Canvas Sets", the AUTH_DISABLED/sign-in warnings, and the never-hand-build-URL/token warnings each appear in multiple sections — state each once in the most relevant section.

Tighten dense run-on paragraphs (Core Model, Account And Sharing Model) into short bullet constraints; much of the trust-boundary reasoning can be compressed to the operative rules without losing safety.

DimensionReasoningScore

Conciseness

The body is information-dense with little explanation of concepts Claude already knows, but at ~600 lines it carries noticeable redundancy that could be tightened: the frame-count guidance is stated twice in "Useful Canvas Sets" ("Do not expand beyond 7 frames..." and "Do not expand every discovered route..."), the sign-in/AUTH_DISABLED and never-hand-build-URL warnings recur across sections, and long run-on security paragraphs (e.g. the Core Model trust discussion) repeat constraints already made elsewhere. Not a 2 because almost every sentence carries skill-specific operational content rather than padding; not a 4 because the duplication and dense multi-clause paragraphs are more than minor.

3 / 5

Actionability

Guidance is fully executable throughout: copy-paste bash for token generation and the durable bridge (AGENT_NATIVE_BRIDGE_TOKEN=... npx @agent-native/core@latest design connect --daemon), complete JSON payloads for open-visual-edit including viewports and resume cases, concrete JS for both the page helper and standard WebMCP, exact tool names (get-visual-edit-pending, propose-node-rewrite, write-local-file), ports, and health-check commands. Examples cover the common cases (fresh signed-out flow, hosted MCP flow, adding frames later, responsive viewports); the only placeholders (routeManifest, rootPath, NAME/ARGS) are appropriately parameterized.

5 / 5

Workflow Clarity

Multi-step flows are explicitly sequenced with validation checkpoints and feedback loops: verify the target URL responds and probe routes before opening Design; numbered bridge steps 1-3 with the mint-then-adopt ordering explained; the /health and manifest probe to distinguish a live bridge from a broken one; the pull -> apply -> acknowledge-revision -> pull-again loop with "Never acknowledge before applying"; version-hash conflict handling ("on conflict, re-read and re-plan"); and explicit failure diagnosis ("If frames do not refresh, the write did not land; say so"). This matches the anchor with explicit validation steps and error-recovery loops.

5 / 5

Progressive Disclosure

Section headers are clear and the one external reference (the GitHub Visual Edit guide) is well signaled, but there are no bundle files at all — no references/, scripts/, or assets/ — so a large amount of specialized content that clearly belongs one level deep is inlined in SKILL.md: per-host browser/WebMCP instructions, the React source-writeback provenance rules, the account/sharing security model, and manual screen-management details. This sits between 'content that should be separate is inline' (3) and 'most content appropriately placed' (4) — the structure is good, but a 600-line body with this much peripheral depth should split most of it into reference files.

3 / 5

Total

16

/

20

Passed

Description

83%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A strong description: concrete capabilities, third-person voice, and an explicit 'Use when' clause with natural trigger verbs. Its main gaps are a few missing synonyms (localhost, dev server, preview) and slightly compressed action coverage.

DimensionReasoningScore

Specificity

The description lists several concrete capabilities — "Open and collaboratively edit a running local app in Design", "shared fallback previews and source handoff" — going beyond naming the domain. It is not a 5 because coverage is thin: core actions like applying edits back to source, route placement, or viewport sets are absent, and "shared fallback previews" is a feature rather than an action.

4 / 5

Completeness

It explicitly answers both questions: what ("Open and collaboratively edit a running local app in Design, with shared fallback previews and source handoff") and when ("Use when the user asks to inspect, share, or edit a real local app in Design") with concrete trigger phrases. Not below 5 because the 'when' clause is explicit and verb-specific rather than implied.

5 / 5

Trigger Term Quality

Natural verbs users would say are present: "inspect", "share", or "edit a real local app". It is not a 5 because common synonyms a user might naturally use — "localhost", "dev server", "preview", "visual edit", "running app" — are missing, leaving a few natural trigger phrases uncovered.

4 / 5

Distinctiveness Conflict Risk

The niche — visual/collaborative editing of a running local app inside Design — is distinct and the triggers are pinned to that domain. Not a 5 because phrases like "inspect" or "edit a real local app" could plausibly fire for generic app-inspection or code-editing requests that are not about the Design canvas.

4 / 5

Total

17

/

20

Passed

Validation

87%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 14 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

skill_md_line_count

SKILL.md is long (609 lines); consider splitting into references/ and linking

Warning

metadata_version

'metadata.version' is missing

Warning

Total

14

/

16

Passed

Repository

BuilderIO/agent-native

Path

.agents/plugins/agent-native-design/skills/visual-edit/SKILL.md

Reviewed

Table of Contents

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.