eve-verification-plans（前夕验证计划）

Eve Verification Plans

Author agentic verification plans — markdown documents that fully specify the steps to verify an Eve-compatible app works correctly AND conforms to Eve platform conventions. Plans are actionable by humans or agents.

When to Use

• Building verification for a new Eve-compatible app
• Auditing an existing app for Eve platform conformance
• After significant feature work that needs structured validation
• Before handoff — proving the app works the Eve way, end to end
• Onboarding a new team to understand what "correctly built on Eve" looks like

Quick Start

# 1. Create the verification directory in your project
mkdir -p e2e-verification/00-smoke

# 2. Copy the smoke template
cp templates/00-smoke-test-plan.md e2e-verification/00-smoke/

# 3. Customize for your app (endpoints, services, agents)
# 4. Run it

The Six Verification Dimensions

Every Eve app has up to six dimensions to verify. Cover all that apply.

Verification Plan Format

Each scenario is a self-contained markdown document:

# Scenario NN: <Name>

**Time:** ~Nm
**Environment:** staging | local | both
**Parallel Safe:** Yes/No
**Requires:** LLM | Browser | None

<one-paragraph description>

## Prerequisites

- What must be true before running
- Required secrets, auth, prior scenarios

## Fixtures

- File paths used by this scenario
- Provenance or generation command for each
- Why these files are representative

## Setup

\```bash
# Environment detection + auth
# Project/org setup
# Fixture validation or generation
\```

## Phases

### Phase 1: <Name>

\```bash
# Commands to execute
\```

**Expected:**
- Bullet list of assertions
- Each assertion is pass/fail verifiable

### Phase 2: ...

## Success Criteria

- [ ] Checkboxes for every pass/fail assertion
- [ ] Grouped by phase

## Debugging

| Symptom | Diagnostic | Fix |
|---------|-----------|-----|
| ... | ... | ... |

## Cleanup

\```bash
# Teardown commands
\```

Format Rules

• Environment-aware: Every plan starts with environment detection — EVE_API_URL determines cloud vs local
• Self-contained: No assumed state beyond documented prerequisites
• Fixture-explicit: Every uploaded/imported artifact is checked in or generated from documented commands
• Phased: Break into phases that can run independently (parallel where safe)
• Assertion-driven: Every step has explicit Expected: with pass/fail criteria
• Debuggable: Troubleshooting section with symptom → diagnostic → fix

Platform Conformance Verification

Before testing functionality, verify the app is built the Eve way. Every verification suite starts with this checklist:

• .eve/manifest.yaml exists and passes eve project sync --dry-run
• Manifest follows current conventions (name preferred over legacy project)
• All services have health endpoints reachable via Eve ingress
• CLI can interact with every API endpoint (no "UI-only" functionality)
• Secrets managed via eve secrets, not hardcoded or env-file-based
• DB migrations run as pipeline steps, not manual scripts
• Agents (if any) defined in agents.yaml with harness profiles
• Pipelines (if any) defined in manifest and runnable via eve pipeline run
• Frontend (if any) authenticates via Eve SSO, not custom auth
• Upload/import flows (if any) have deterministic fixtures checked in or generated locally

See references/eve-conformance-checks.md for the full checklist with rationale.

Service Layer Verification

Test every API surface CLI-first:

TOKEN=$(eve auth token --raw)

# Health check (always first)
curl -sf "${APP_SCHEME}://api.${APP_DOMAIN}/health" | jq '.'

# App API via auth token
curl -sf -H "Authorization: Bearer $TOKEN" \
  "${APP_SCHEME}://api.${APP_DOMAIN}/endpoint" | jq '.field'

CLI parity assertion: For every curl call in a test plan, ask: "Can this also be done via a CLI command?" If not, file an issue — don't accept it.

Pattern:

1. Eve CLI commands where they exist (deploy, env, job, secrets)
2. App CLI if the app follows eve-app-cli patterns
3. REST API via curl with auth tokens for custom endpoints
4. Auth: Mint tokens via eve auth mint or eve auth token --raw

Input / Ingestion Verification

When the app accepts uploads, imports, or document bundles, verification must include a fixture plan.

Fixture Selection Order

1. Reuse existing repo fixtures if they match accepted file types and are deterministic
2. Manufacture synthetic fixtures locally with committed scripts
3. Source small public-domain fixtures only when local manufacture would reduce realism

Fixture Matrix

• Minimal valid — smallest acceptable file that exercises the happy path
• Typical real-world — representative document/media/import file
• Boundary / invalid — wrong type, malformed structure, or size edge
• Cross-format — if the app accepts multiple types (PDF + Markdown + CSV), verify each

What to Check

• File accepted through the real app surface (CLI, REST endpoint, or browser upload)
• MIME/type detection and metadata are correct
• Storage/persistence path is correct
• Downstream processing produces expected results
• Error handling is explicit for rejected or malformed fixtures

Rule: If a plan says "upload a sample PDF", it must include an actual file path or generator step. "Find a PDF online" is not acceptable.

See references/fixture-patterns.md for detailed guidance.

UI Verification

When the app has a frontend, verify visual quality and interaction flows.

SSO Token Injection

# Mint an SSO token via CLI
SSO_TOKEN=$(eve auth mint --email user@example.com --org $ORG_ID --format sso-jwt)

# Use agent-browser with the token
agent-browser --session verify open "${APP_URL}/auth/callback?token=${SSO_TOKEN}"
agent-browser --session verify wait --url "**/dashboard"
agent-browser --session verify screenshot ./e2e-verification/artifacts/dashboard.png

What to Check

• Pages render without console errors
• Dark mode and light mode both work (screenshot both)
• Key user flows complete (login → dashboard → action → result)
• Responsive layout at standard breakpoints
• Forms submit correctly and validation fires

See references/ui-verification-patterns.md for tool choice guidance and patterns.

Agent Verification

When the app includes Eve agents, verification extends to behavior quality.

1. Create a job that exercises the agent's primary workflow
2. Follow execution: eve job follow <job-id>
3. Check receipt: eve job receipt <job-id> (tokens, cost, duration)
4. Apply eve-agent-optimisation diagnostic workflow
5. Record baseline metrics in the test plan for regression detection

What to Check

• Agent completes its task (correct output)
• Token usage within acceptable bounds
• No unnecessary tool calls or blind alleys
• Error cases handled gracefully (bad input, missing secrets)
• Multi-agent coordination works (jobs complete in dependency order)

See references/agent-verification-patterns.md for integration with optimization.

Deploy Cycle Patterns

Verification often reveals issues. The fix/deploy loop:

Cloud (Staging) — Default

discover bug → fix code → commit → tag release-v* → push tag →
  wait for CI (publish-images → infra dispatch → deploy) →
  re-run failed scenario

Local (k3d)

discover bug → fix code → pnpm build →
  ./bin/eh k8s-image push → ./bin/eh k8s deploy →
  re-run failed scenario

See references/deploy-cycle-patterns.md for environment detection and wait patterns.

Scenario Discovery

How to identify what scenarios to create for a given app:

1. Read the manifest — every service is a verification target
2. Read agents config — every agent needs a behavioral test
3. Read the API spec — every endpoint group is a potential scenario
4. Check upload/import surfaces — every accepted file class needs fixtures
5. Check pipelines — build/deploy/workflow pipelines need end-to-end verification
6. Check the UI — every page/route needs visual verification
7. Check integrations — webhooks, chat gateways, external APIs

Minimum Scenario Set

Running Verification Plans

Sequential Execution

# Run scenarios in order
for plan in e2e-verification/*/; do
  echo "=== Running: $plan ==="
  # Agent reads and executes the plan document
done

Parallel Execution

Scenarios marked Parallel Safe: Yes can run concurrently. Typically:

• 00-smoke runs first (validates prerequisites)
• 01-deploy through 03-ui-visual can parallelize
• Agent scenarios (06-07) depend on deploy completing

CI Integration

# In CI, set environment and run
export EVE_API_URL=https://api.eh1.incept5.dev
eve auth login --email $CI_EMAIL --ssh-key $CI_SSH_KEY

# Run all scenarios, collect artifacts
mkdir -p e2e-verification/artifacts
# Agent executes each plan, screenshots/logs go to artifacts/

File Structure

<project-root>/
  e2e-verification/
    README.md                    # Index of all scenarios
    00-smoke/
      00-smoke-test-plan.md
    01-deploy/
      01-deploy-test-plan.md
    02-core-flow/
      02-core-flow-test-plan.md
      fixtures/
        README.md
        test-data.json
    03-ui-visual/
      03-ui-visual-test-plan.md
    04-input-ingestion/
      04-input-ingestion-test-plan.md
      fixtures/
        README.md
        sample-document.pdf
        sample-import.csv
        scripts/
          make-fixtures.sh
    artifacts/                   # Generated during runs (gitignored)

Numbering: NN-kebab-name/ directories with matching NN-kebab-name-test-plan.md. Numbers imply execution order. Gaps are fine — they allow insertion without renaming.

Fixtures: Every scenario that depends on uploaded or imported inputs gets a sibling fixtures/ directory. fixtures/README.md records provenance, generation steps, and sensitivity notes.

References

• references/test-plan-format.md — Full format specification with annotated examples
• references/eve-conformance-checks.md — The Eve way: what to verify and why
• references/fixture-patterns.md — Fixture sourcing, manufacture, and documentation
• references/deploy-cycle-patterns.md — Fix/deploy loop for cloud and local
• references/ui-verification-patterns.md — Browser automation and SSO auth patterns
• references/agent-verification-patterns.md — Agent testing and optimization integration

Templates

• templates/00-smoke-test-plan.md — Starter smoke + conformance template
• templates/scenario-test-plan.md — General scenario skeleton
• templates/upload-ingest-test-plan.md — Input-heavy scenario with fixture matrix

Related Skills