Skip to content

Pipeline Modes

Comparison reference for all Quorvex AI execution pipelines.

Pipeline Comparison

Mode CLI Flag Stages Healing Attempts Browser Required
Default (none) Plan + Generate + Heal 3 (native) Yes
Hybrid --hybrid Plan + Generate + Heal 3 (native) + 17 (Ralph) Yes
Standard (legacy) --standard-pipeline Plan + Operator + Export + Validate 7 Yes
PRD --prd PDF Extract + Plan + Generate + Heal 3 (native) Yes
Exploration --explore URL AI Discovery N/A Yes
Skill --skill-mode Script Generation + Execution N/A Yes

Pipeline

Property Value
Default Yes
CLI flag (none)
Stages Planner, Generator, Healer
Healing attempts 3
Healing tool Playwright test_debug MCP
Browser interaction Every stage

Stage Sequence

Stage Description
1. Planning AI agent opens browser, navigates to target URL, explores page, builds plan with discovered selectors
2. Generation AI agent opens fresh browser, reads spec and plan, writes Playwright test code with live app interaction
3. Execution Generated test runs via npx playwright test
4. Healing On failure: healer uses test_debug MCP tool, rewrites failing portions, re-runs (up to 3 attempts)

Hybrid Pipeline

Property Value
Default No
CLI flag --hybrid
Stages Default pipeline + Ralph escalation
Total healing budget 20 (configurable via --max-iterations)
Phase 1 (native) Attempts 1-3
Phase 2 (Ralph) Attempts 4-20

Standard Pipeline (Legacy)

Property Value
Default No
CLI flag --standard-pipeline
Stages Planner, Operator, Exporter, Validator
Healing attempts 7
Browser interaction Operator stage only
Status Not recommended for new tests

PRD Pipeline

Property Value
Default No
CLI flag --prd
Input PDF file
Stages PDF extraction, spec generation (per feature), native pipeline (per spec)
Feature selection --feature "Feature Name"
Spec splitting --split

Exploration Pipeline

Property Value
Default No
CLI flag --explore URL
Output Discovered pages, flows, API endpoints
Strategies goal_directed (default), breadth_first, depth_first
Max interactions --max-interactions (default: 50)
Timeout --timeout (default: 30 minutes)
Authentication --login-url URL + LOGIN_EMAIL/LOGIN_PASSWORD env vars

Exploration Strategies

Strategy Behavior
goal_directed Prioritizes meaningful user flows and actions
breadth_first Visits as many pages as possible before going deep
depth_first Follows each flow to completion before moving to the next

Skill Pipeline

Property Value
Default No
CLI flag --skill-mode (from spec) or --run-skill SCRIPT (direct)
Output JavaScript skill script
Timeout --skill-timeout (default: 30000 ms)
Headless --skill-headless

Smart Check (Stage 0)

Runs before any pipeline. Checks for existing generated code.

Condition Action
Valid code exists and passes Skip pipeline, reuse code
Code exists but fails Jump to healing (skip plan + generate)
No code exists or healing fails Run full pipeline
--try-code PATH specified Use the specified file as existing code

Common Flags

Flag Type Default Applies To
--browser chromium / firefox / webkit chromium All modes
--project-id string derived from spec folder All modes
--no-memory flag false All modes
--run-dir path runs/TIMESTAMP/ All modes
--interactive / -i flag false Standard pipeline only
--max-iterations int 20 Hybrid mode

Decision Matrix

Scenario Recommended Mode
New test from a simple spec Default
Test keeps failing after 3 healing attempts Hybrid (--hybrid)
Generate tests from a PDF PRD PRD (--prd)
Discover what an app does Exploration (--explore)
Need network mocking or multi-tab Skill (--skill-mode)
Reproduce a legacy run Standard (--standard-pipeline)
Test on Firefox or WebKit Any mode + --browser firefox
Disable memory system Any mode + --no-memory