looplia-e2e

name	looplia-e2e
description	End-to-end testing for looplia CLI from local source. Builds the CLI, initializes workspace, runs the writing-kit workflow, and verifies outputs. Use when testing local development or validating workflow execution.
license	MIT
compatibility	Requires jq and bun. Works with Claude Code.
metadata	{"author":"looplia","version":"0.8.0"}

Looplia E2E Test Skill

End-to-end testing for looplia CLI from local source.

Quick Start

# Set API key in .env
echo "ZENMUX_API_KEY=your-key" >> .env

# Run E2E test
.claude/skills/looplia-e2e/scripts/e2e.sh

Running inside Claude Code? Use env -u CLAUDECODE to allow the CLI to spawn a nested Claude Code subprocess. The CLAUDECODE env var is set by the outer session and blocks any nested claude invocations:

env -u CLAUDECODE .claude/skills/looplia-e2e/scripts/e2e.sh

The scripts internally unset CLAUDECODE in setup_test_env(), so this is only needed if you call e2e.sh directly (not via the looplia-e2e skill invocation).

Modular Test Scripts (v0.8.0)

The E2E tests are modular and can be run individually:

Run All Tests

.claude/skills/looplia-e2e/scripts/e2e.sh

Run Specific Tests

# Auto-discovery only
./scripts/e2e.sh --test auto

# Workflow execution only
./scripts/e2e.sh --test workflow

# Or run directly
./scripts/e2e-auto-discovery.sh

Test Isolation

Tests use LOOPLIA_HOME to isolate from your real workspace:

• Test workspace: <project>/test-workspace/
• Your workspace: ~/.looplia/ (untouched)

What It Tests

The script performs these steps:

1. Build - Compiles the CLI from source
2. Reset - Removes test workspace for fresh start
3. Init - Initializes workspace with plugins
4. Configure - Sets provider to ZenMux MiniMax M2.5
5. Build Command - Tests workflow generation with auto-discovery (HN AI news aggregator)
6. Run Command - Executes writing-kit workflow with ai-healthcare.md
7. Verify - Checks outputs, validation state, and auto-discovery results

Expected Outputs

Build command:

test-workspace/workflows/e2e-auto-discovery-test.md    # Generated workflow file
test-workspace/sandbox/build-<id>/
├── validation.json       # workflowValidated: true
└── logs/
    └── *.log             # Execution logs
test-workspace/plugins/auto-discovery-plugin/
├── .claude-plugin/       # Plugin initialized
└── skills/               # Auto-discovered skills

Run command:

test-workspace/sandbox/<run-id>/
├── outputs/
│   ├── summary.json      # Stage 1: Content analysis
│   ├── ideas.json        # Stage 2: Idea generation
│   └── writing-kit.json  # Stage 3: Final output
├── validation.json       # All steps validated: true
└── logs/
    └── *.log             # Execution logs

Success Criteria

Auto-discovery test:

• Workflow file created at test-workspace/workflows/e2e-auto-discovery-test.md
• validation.json shows workflowValidated: true
• Auto-discovery plugin initialized
• Skills discovered and cataloged (external dependency)

Workflow test:

• 3 output files created (summary.json, ideas.json, writing-kit.json)
• All 3 steps validated in validation.json
• Final output writing-kit.json exists

Troubleshooting

Transient API errors (retry usually works): The ZenMux provider may occasionally return transient errors like "duplicate tool_call id". This is a provider-side issue, not a workflow bug. Simply run the test again:

# Just retry - second run usually succeeds
.claude/skills/looplia-e2e/scripts/e2e.sh

API key issues:

# Verify .env exists and contains ZENMUX_API_KEY
cat .env | grep ZENMUX_API_KEY

Workspace issues:

# Manual reset (test workspace only)
rm -rf test-workspace && looplia init --yes

Build issues:

# Clean rebuild
rm -rf apps/cli/dist && bun run build

File Structure

.claude/skills/looplia-e2e/
├── SKILL.md                    # This file
├── scripts/
│   ├── e2e-setup.sh           # Shared setup/cleanup
│   ├── e2e-auto-discovery.sh  # Auto-discovery focused test
│   └── e2e.sh                 # Orchestrator (runs all tests)
└── assets/
    └── ai-healthcare.md       # Test content fixture