界面已本地化;技能介绍与说明文档按所选语言由 AI 自动翻译,代码、命令、路径和专业术语保留原文,以原作者文档为准。
探索技能
数据与研究 · SKILL.md

lucidchart-debug-bundle

lucidchart-debug-bundle

【数据与研究】Assemble a redacted, reproducible Lucid integration diagnostic bundle. Use when an API, Standard Import, extension, export, or data connector failure needs escalation. Trigger with "build Lucid debug bundle".

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: lucidchart-debug-bundle description: 'Assemble a redacted, reproducible Lucid integration diagnostic bundle. Use when an API, Standard Import, extension, export, or data connector failure needs escalation. Trigger with "build Lucid debug bundle".' argument-hint: "[project-path] [incident-id]" allowed-tools: Read, Glob, Grep, WebFetch, Write, Edit version: 1.8.0 author: Jeremy Longshore jeremy@intentsolutions.io license: MIT tags: [saas, lucidchart, debugging, diagnostics, support] model: inherit effort: medium compatibility: Designed for Claude Code; bundle sharing requires data-owner approval and a secure approved support channel

Redacted Lucid Debug Bundle

Overview

Create a minimal evidence package that reproduces a Lucid failure while excluding credentials, sensitive document content, and unnecessary personal data.

Prerequisites

  • Incident identifier, time window, affected component, and expected behavior
  • Approval to inspect relevant local artifacts and sanitized responses
  • A clean output directory outside application source and secret stores

Tool Discipline

Use Read, Glob, and Grep to collect bounded evidence, WebFetch for current contract comparison, and Write or Edit only inside the declared bundle directory.

Current Contract

Useful evidence differs by surface: REST requests need method, path family, safe headers, status and request identifiers; Standard Import needs archive structure and sanitized samples; extensions/connectors need versions, manifests, build output, and minimal fixtures.

Authentication

Never copy Authorization, cookies, API keys, OAuth codes, refresh tokens, client secrets, signed URLs, or raw environment files. Replace sensitive identifiers consistently so relationships remain diagnosable.

Instructions

  1. Declare incident scope, bundle path, retention, recipients, and exclusion rules.
  2. Inventory candidate files and reject caches, dependencies, binaries, full documents, and unrelated logs by default.
  3. Capture tool and SDK versions, safe manifest fields, operation category, sanitized request/response metadata, and exact timestamps.
  4. For imports, include a structurally equivalent minimal archive or manifest—not protected production content.
  5. For extension/connector failures, add a synthetic fixture and the smallest failing build/test receipt.
  6. Scan recursively for token patterns, credentials, emails, personal data, and document content; manually review hits.
  7. Write a manifest with file digests, provenance, redactions, reproduction steps, expected/actual behavior, and gaps.
  8. Obtain approval before transmitting; record destination and deletion date.

Approval Boundaries

Do not collect from production, contact Lucid support, or transmit the bundle without incident and data-owner authorization.

Output

Return bundle path, manifest digest, files included/excluded, redaction results, reproduction status, evidence gaps, and sharing approval.

Error Handling

ConditionResponse
Secret scanner reports a hitRemove or irreversibly redact it, then rescan the entire bundle.
Failure cannot be reproducedPreserve timestamps and safe request identifiers; label hypotheses.
Minimal fixture still contains sensitive dataReplace it with synthetic structure before sharing.

Example

incident=LUCID-42; files=8; secrets=0; pii=0; reproduction=pass; manifest-sha256=...; shared=no

Resources

Next Steps

Share only through the approved support channel, then enforce the recorded retention deadline.

我的收藏