| name | manual-testing-guide |
| description | Generate a comprehensive manual testing guide for completed work, with note sections for user feedback that the agent reviews afterward. Use when the user wants to manually test changes before shipping. |
Manual Testing Guide
Generate a thorough manual testing guide for the work just completed. The guide is a markdown file the user fills out during testing, then hands back for review.
Workflow
Phase 1: Generate the Guide
- Analyze the changes — run
git diff (staged + unstaged) and review all modified/added files to understand exactly what changed.
- Identify test surfaces — map changes to user-facing behavior, API endpoints, CLI commands, UI interactions, config changes, and side effects.
- Write the guide — create
TESTING.md in the repo root using the format below.
Phase 2: Review the Notes
When the user says testing is done (or asks you to review), read TESTING.md and:
- Flag any test case marked
[x] FAIL or with concerning notes.
- Suggest fixes or next steps for failures.
- Identify any gaps — test cases the user skipped or scenarios the notes reveal were missed.
- Summarize overall readiness: ship, fix-then-ship, or needs-more-work.
Guide Format
# Manual Testing Guide
**Changes**: (one-line summary of what was changed)
**Date**: YYYY-MM-DD
**Branch**: (branch name)
---
## Prerequisites
- (environment setup, seed data, config needed before testing)
---
## Test Cases
### 1. (Test case title)
**Category**: (Happy Path | Edge Case | Error Handling | Regression | Security | Performance)
**Priority**: (P0 — must pass | P1 — should pass | P2 — nice to verify)
**Steps**:
1. Step one
2. Step two
3. Step three
**Expected Result**: (what should happen)
**Result**: [ ] PASS [ ] FAIL
**Notes**:
> (tester writes observations, screenshots, unexpected behavior here)
---
(repeat for each test case)
---
## Summary
**Overall Result**: [ ] PASS [ ] FAIL
**Tested By**:
**General Notes**:
> (overall observations, concerns, or follow-ups)
Test Case Coverage Rules
Generate test cases for every category that applies:
- Happy Path — the primary intended flow works correctly
- Edge Cases — boundary values, empty inputs, max lengths, special characters, concurrent usage
- Error Handling — invalid inputs, missing data, network failures, permission errors
- Regression — existing functionality that touches the same code still works
- Security — auth checks, input sanitization, privilege escalation (if applicable)
- Performance — large datasets, rapid repeated actions (if applicable)
Order test cases by priority (P0 first), then by category.
Guidelines
- Be specific in steps — include exact inputs, URLs, commands, or clicks. Never say "test the feature."
- Expected results must be observable and verifiable, not vague ("it works").
- Keep the guide practical — 10-25 test cases is typical. Don't pad with trivial cases.
- If changes span multiple features, group test cases under feature subheadings.
- Include a Prerequisites section only if setup is actually needed.
- Write for someone who didn't make the changes — enough context to test without reading the code.