Skip to content
Back to skills

Ramp Debug Bundle

ASecurity

Collect a minimal, safe Ramp API diagnostic bundle centered on trace IDs and reproducible contract evidence. Use when escalating a persistent API, webhook, deferred-task, or accounting failure. Trigger with "Ramp debug bundle" or "Ramp support evidence".

  • 2,719 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added September 21, 2026
ai-agentsdebuggingapisecuritydocumentation

Works with

  • cli
  • api

Security analysis

A100/100

Pro scans all 2 files and shows the line behind each finding

Scanned September 21, 2026

npx -y skills add jeremylongshore/claude-code-plugins-plus --skill ramp-debug-bundle --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Ramp Debug Bundle?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Ramp Debug Bundle
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/jeremylongshore-ramp-debug-bundle-bd0331f6/badge)](https://www.skillsdirectory.com/skills/jeremylongshore-ramp-debug-bundle-bd0331f6)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: ramp-debug-bundle
description: >-
  Collect a minimal, safe Ramp API diagnostic bundle centered on trace IDs and reproducible contract evidence. Use when escalating a persistent API, webhook, deferred-task, or accounting failure. Trigger with "Ramp debug bundle" or "Ramp support evidence".
allowed-tools: Read,Glob,Grep,Write,Edit
version: 2.0.0
argument-hint: "[integration-or-environment]"
model: inherit
effort: high
author: Jeremy Longshore <jeremy@intentsolutions.io>
license: MIT
compatibility: Requires current Ramp Developer API documentation and approved access for any live financial, card, identity, accounting, application, or configuration change
tags: [saas, ramp, debugging, support, evidence]
---
# Sanitized Ramp Support Bundle

## Overview

Create an evidence package that Ramp support and internal responders can use without exposing credentials, card details, receipts, or unnecessary employee and financial data.

## Prerequisites

- Identify the Ramp application, environment, business entities, affected data and workflows, accountable owner, and rollback boundary.
- Read `references/official-docs.md` and re-check endpoint schemas, scopes, limits, and support status before a live operation.
- Use synthetic fixtures or Ramp sandbox until production access and business effects are explicitly approved.
- Prepare approved secret storage and a sanitized evidence location.

## Current Contract

- Ramp responses include `x-trace-id`, the preferred correlation handle for developer support.
- API error bodies may include `error_v2.error_code`, additional information, and a message.
- Webhook and deferred-task incidents need delivery/task identifiers and lifecycle state, not copied secrets or full payloads.
- Authorization failures require environment, grant, and scope metadata, but never the bearer token or client secret.

## Instructions

1. Open a restricted case directory and record incident window, environment, affected endpoint or event, expected behavior, owner, and data-handling classification.

2. Capture UTC timestamps, method, sanitized path template, status, `x-trace-id`, error code, attempt count, latency, idempotency key hash, and application release.

3. Add a minimized schema-valid reproducer using placeholders and sandbox IDs; record grant type and scope names without credential values.

4. For webhooks or deferred tasks, include subscription/event/task identifiers, deduplication outcome, state transitions, and sanitized response metadata.

5. Run a secret/card/identity scan, obtain the data owner's sharing approval, generate checksums, and transmit only through the approved support channel.

## Tool Discipline

- Use **Glob** to locate candidate code, manifests, fixtures, and evidence without widening scope.
- Use **Grep** to find relevant endpoints, fields, permissions, identifiers, errors, and stale assumptions.
- Use **Read** to inspect the smallest required local files and authoritative evidence.
- Use **Write** only for a new approved local draft, test, configuration, or evidence artifact.
- Use **Edit** only for a bounded approved change with a known rollback.
- Local file tools do not authorize a Ramp operation or replace owner approval.

## Approval Boundaries

The incident owner approves scope; security/data owners approve external sharing. Never upload raw production bodies, receipts, PAN/CVV, tokens, or client secrets.

## Output

A checksummed sanitized bundle, field manifest, reproduction steps, trace IDs, environment and release metadata, redaction scan, sharing approval, and retention deadline.

## Error Handling

| Condition | Response |
|---|---|
| No trace ID was retained | Reproduce the smallest safe request in sandbox or an approved read-only path and capture the new trace. |
| Redaction destroys reproducibility | Replace values with type-preserving placeholders and keep restricted originals outside the shared bundle. |
| A support request asks for secrets | Refuse and use an approved credential-verification or secure escalation process. |

## Examples

### Example 1

Bundle a recurring 422 with one sanitized request fixture, `error_v2` code, trace ID, schema version, and exact release.

### Example 2

Document a webhook gap using event IDs, receive timestamps, dedupe decisions, and a reconciled object count without event bodies.

## Validation

- The bundle reproduces or precisely bounds the failure.
- Trace IDs and identifiers are preserved while secret and sensitive value scans are clean.
- Every included field has an explicit diagnostic purpose.
- Sharing channel, approver, checksum, and deletion date are recorded.

## Resources

- [Official documentation and contract notes](references/official-docs.md)
- Re-check the dated contract and current OpenAPI schema before any live request.
- Treat unresolved vendor behavior, authority, or financial state as a stop condition.

Files in this skill

  • SKILL.md4.8 KB
  • references/official-docs.md820 B

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.