Kheprilab
DocsWorkflows
Using wep

Workflows

Choosing and combining working methods: fast-patch, research-first, SDD, TDD, EDD and their combinations.

wep supports explicit workflows without forcing every task into the same process.

Supported Workflows

  • fast-patch
  • research-first
  • SDD
  • TDD
  • EDD
  • SDD+TDD
  • SDD+EDD
  • TDD+EDD
  • UI review (auto-assigned via workflowRules.ui, not user-selectable as a default)

Use fast-patch for small, low-risk changes with obvious behavior.

Use research-first when the code area is unknown or the architecture is unclear.

Use SDD for APIs, contracts, schemas, permissions, and externally visible behavior.

Use TDD for bugs, regressions, and logic changes with clear expected outcomes.

Use EDD when examples and edge cases are the main source of clarity.

Use SDD+TDD for new or changed contracts that need executable tests.

Use SDD+EDD for rule-heavy domains where examples and boundaries matter.

Use TDD+EDD for bug fixes where examples clarify edge cases.

When using the optional implementer with a TDD workflow, its implementation log must preserve one ordered red→green pair in the same round. The red validation records phase: "red", status: "failed", and an expectedFailure whose text appears in its verbatim output; the green validation records phase: "green", status: "passed", and follows with the same command after whitespace normalization. A missing or mismatched pair blocks the implementer handoff. Older logs remain readable but do not receive invented evidence.

UI review is a task-level automatic recommendation, not a valid --workflow argument or configurable default. Use it for tasks that require reviewing or validating UI quality. The agent discovers available tools (stylelint, eslint, storybook, playwright, chromatic, etc.) and runs them without modifying code. If the project defines its own UI review conventions, those take precedence.

Agent-First Flow

After wep init, the selected coding agent should own task preparation:

Run wep task workflow for invoice API examples. Use SDD+EDD.

The CLI task command is a deterministic fallback:

wep task "Define invoice API examples" --workflow SDD+EDD

Every task capsule should record:

## Recommended Workflow

SDD+EDD

## Workflow Notes

- Workflow source: explicit user request.
- Why this workflow: invoice behavior needs a contract plus edge-case examples.

Selection rules and limits

The coding agent reads defaults.workflow, defaults.allowedWorkflows, and workflowRules from the project config. These are instructions for a cooperative agent. The CLI fallback uses a narrower deterministic algorithm:

  1. An explicit --workflow wins, provided it is in allowedWorkflows.
  2. Otherwise, a concrete defaults.workflow wins and must be allowed.
  3. With task-fit, hard-coded English keywords recommend a workflow. The CLI does not execute the workflowRules object as a rule engine. UI terms can recommend UI review; other disallowed recommendations fall back to the first allowed workflow.

UI review is not filtered by allowedWorkflows. For example, wep task "Review settings screen" can recommend it, but wep task "Review settings screen" --workflow "UI review" is rejected. A UI review capsule is for inspection; ask for a suitable implementation workflow when building or changing a UI.

The CLI creates a skeleton with placeholder context and criteria. It does not inspect the codebase, resolve ambiguity, generate tests, or enforce the agent’s execution of SDD, TDD, or EDD. Complete the capsule before using it for implementation or review.