Skip to main content
ClaudeWave
Skill2.1k repo starsupdated 17d ago

superglue-integration-builder

Build and verify Superglue natural-language integrations. Use this skill when an agent is asked to plan, build, review, or verify Superglue workflows and needs plugin-specific acceptance criteria.

Install in Claude Code
Copy
git clone --depth 1 https://github.com/superglue-ai/superglue /tmp/superglue-integration-builder && cp -r /tmp/superglue-integration-builder/plugins/superglue-integration-builder/skills/superglue-integration-builder ~/.claude/skills/superglue-integration-builder
Then start a new Claude Code session; the skill loads automatically.

SKILL.md

# Superglue Agent Plugin

This skill turns a broad agent request into a Superglue-specific workflow with explicit verification evidence. It is intentionally operational: it should produce a plan, the expected artifacts, and plugin eval metadata that another maintainer can review.

## Capabilities

- Schema discovery.
- Integration mapping.
- Auth setup review.
- Generated tool qa.

## Workflow

1. Api/schema discovery.
2. Connector mapping review.
3. Auth boundary check.
4. Integration regression pack.

## Required Output

Return a concise implementation or review note with these sections:

- `Scope`: the exact Superglue workflow, repository area, and user-facing outcome.
- `Inputs`: non-secret configuration, sample IDs, file paths, docs, or local commands needed to proceed.
- `Plan`: ordered steps the agent should take, including where human approval is required.
- `Verification`: commands, UI checks, fixtures, screenshots, traces, or logs that prove the plugin workflow behaved correctly.
- `Plugin Eval Metadata`: the eval case id, expected pass criteria, and any safe metadata events to record.
- `Risks`: unresolved assumptions, missing credentials, destructive operations, or compatibility concerns.

## Acceptance Checks

- Lists source and target systems.
- Documents auth type without recording credentials.
- Maps fields with required/optional status.
- Proposes replay-safe regression examples.

## Privacy And Telemetry Boundary

Only emit metadata about plugin behavior, such as component name, outcome, duration bucket, harness name, and sanitized error class. Do not emit prompts, file contents, connector payloads, API tokens, request bodies, model outputs, user data, or production identifiers.

## Optional Telvine Measurement

Teams that publish this plugin through Telvine can measure adoption and eval outcomes without changing Superglue runtime code. The CLI package is published as `telvine` on npm: https://www.npmjs.com/package/telvine

```bash
npm i -g telvine
telvine login
telvine publish ./plugins/superglue-integration-builder
telvine plugins metrics
```

## Manifest Path Note

The plugin manifests use `skills: ./skills/` because the supported plugin manifest contract resolves that path from the plugin root, not from the hidden manifest directory.