Delegate menial lookups to Haiku (cost control)
During this skill, push pure read-only lookups DOWN to a cheap subagent instead of running them on the current model. This covers: multi-file grep/glob, "where is X defined / what calls Y", mapping a directory, reading many files to locate something, or fetching a URL for reference.
Use the Agent tool with the caveman:cavecrew-investigator subagent (Haiku,
returns a compressed file:line answer). If that subagent is unavailable, spawn a
general agent with model: haiku. Keep all reasoning, decisions, and edits on
the current model. Delegate only the menial searching.
To E2E Tasks
Analyze the current branch's changes, discover which critical user-facing workflows are affected, stress-test the test plan with the user, and output a Playwright e2e tasks JSON file to docs/tasks/ that /build-code can execute. This skill is a planner — it produces a task list, not implemented test code.
Contract
Format: task file — see contracts/task-contract.md (schema_version: "1")
Role: producer
Step-0 — validate output after writing:
bash ~/.dotfiles/claude-code-shared/scripts/shared/validate-schema.sh \
--instance ~/.dotfiles/claude-code-shared/contracts/task-schema.json \
<output-path>
On non-zero exit: STOP. Report stderr to the user. Do not write the file.
Process
1. Detect Playwright setup
Scan the project for existing Playwright configuration:
find . -maxdepth 3 -name "playwright.config.*" -o -name "playwright.config.*"
If config exists: read it to learn:
- Test directory (e.g.
tests/e2e/,e2e/) - Base URL
- Project names (chromium, firefox, etc.)
- Existing fixtures, helpers, or page objects (scan the test directory)
Record these for use in task descriptions so generated tests follow existing patterns.
If no config exists: the first task in the output JSON will be "Scaffold Playwright setup" with acceptance criteria for config, directory structure, and a sample test. All other tasks will be blocked by this one.
2. Read context
Fire all of the following in the same turn as the step 1 Playwright config scan (parallel — all are independent):
a. Project vocabulary and ADRs. Spawn the context-loader agent (subagent_type: context-loader, repo root as working directory).
b. The diff and commit log (two Bash calls in the same turn):
git diff $(git merge-base HEAD origin/main)..HEAD
git log $(git merge-base HEAD origin/main)..HEAD --oneline
Wait for context-loader, diff, and commit log results before step 2c.
c. The source artifact. Look for an existing task JSON in docs/tasks/ for the current branch. If found, read its source field:
- When
source.typeis"seed"or"prd"andsource.refis non-null: load the artifact atsource.reffor context (seed or PRD respectively). - When
source.typeis"session"(ref null): fall through to the prompt-or-skip path below. - When multiple task files exist: ask the user which one corresponds to the current work.
If no task JSON exists, or the source cannot be resolved, ask the user to provide the seed/PRD path directly or skip context.
Use vocabulary terms from context-loader in workflow names and test scenario descriptions. If the payload's missing list is non-empty, proceed without domain vocabulary.
3. Two-pass workflow discovery
Pass 1: Scan for user-facing signals
Analyze the diff for changes that affect user-facing behavior:
- Route definitions or page components (new/modified)
- API endpoint handlers
- Form handlers, validation logic
- Authentication or authorization changes
- Payment, billing, or subscription logic
- Data mutations visible to users (CRUD operations)
- Navigation changes, redirects
- Error states shown to users
Ignore purely internal changes: utility refactors, type-only changes, build config, dev tooling.
Pass 2: Group into workflows
Cluster the signals from pass 1 into coherent user-facing workflows. Each workflow is a sequence of user actions that achieves a goal.
Examples:
- "User signup flow" (visit signup page, fill form, submit, verify email, land on dashboard)
- "Invoice creation flow" (navigate to invoices, click create, fill details, save, see confirmation)
- "Settings update flow" (open settings, change value, save, see confirmation)
4. Present discovered workflows
Show the user what you found. For each workflow:
- Workflow name: descriptive, using
vocabularyterms from the context-loader payload - Signals found: which diff changes map to this workflow
- Proposed test scenarios: happy path + key edge cases
Discovered workflows from branch changes:
1. Invoice creation flow
Signals: new route /invoices/new, InvoiceForm component, createInvoice API handler
Proposed tests:
- Happy path: create invoice with valid data
- Validation: submit with missing required fields
- Edge: create invoice with maximum line items
2. Invoice PDF export
Signals: new exportPdf handler, PdfPreview component
Proposed tests:
- Happy path: export existing invoice as PDF
- Edge: export invoice with special characters in title
Ask:
- Are these the right workflows to cover?
- Any workflows missing?
- Are the test scenarios sufficient? Too many? Too few?
- Any edge cases I should add or remove?
Iterate until the user approves.
5. Grill the test plan
Once workflows are approved, stress-test the test plan. Apply the same rigor as /grill-with-docs but scoped to the test plan:
Challenge against the glossary. If the context-loader payload returned vocabulary terms, verify test names and descriptions use those canonical domain terms. "Your test says 'checkout' but the project vocabulary calls it 'order submission'. Which is right?"
Sharpen fuzzy scenarios. "You have 'test error handling' but which error? Network timeout? Validation failure? Auth expiry? Be specific."
Probe edge cases. Invent scenarios that test boundaries. "What happens if the user double-clicks submit? What if the session expires mid-flow? What about concurrent edits?"
Cross-reference PRD. If PRD mentions acceptance criteria not covered by any test scenario, surface the gap. "PRD says 'user sees error toast on failed payment' but no test covers the failed payment path."
Cross-reference diff. If the diff contains error handling, branching logic, or state transitions not covered by a test, flag it. "The diff shows a retry loop in the API handler. No test covers the retry path."
Update CONTEXT.md inline if new terms crystallize during the grill. Use the format in ~/.dotfiles/claude-code-shared/skills/grill-with-docs/resources/CONTEXT-FORMAT.md.
Ask questions one at a time. Provide your recommended answer for each. Iterate until the test plan is solid.
6. Determine task structure
After the grill, build the task list:
First task (if page objects/fixtures needed): "Create page objects and fixtures for {workflow names}." This task sets up reusable helpers: page object classes for pages touched across workflows, shared fixtures for auth state, test data setup, etc. Reference existing patterns found in step 1.
Remaining tasks: One task per test scenario. Each task:
- Is type
AFK - Is
blocked_bythe fixtures task only (or nothing if no fixtures task) - Has no dependencies on other test tasks
- Has acceptance criteria describing expected test behavior
- Embeds page objects, fixtures, and workflow context directly in
descriptionso/tddreceives full context (e.g. "Use InvoiceFormPage and InvoiceDetailPage page objects. Setup fixtures: authenticated-user, test-organization. Workflow: invoice creation happy path.")
7. Determine the next task ID
Run ~/.dotfiles/claude-code-shared/scripts/scaffolding/next-task-id.sh docs/tasks/ to get the next available ID.
8. Confirm output directory
Resolve the absolute path of docs/tasks/ relative to the current working directory. Ask the user before writing:
E2E tasks file will be saved to: /absolute/path/to/docs/tasks/
Is that correct?
Use whatever path the user confirms. Create it if it doesn't exist.
9. Confirm branching strategy
MANDATORY. Follow ~/.dotfiles/claude-code-shared/resources/branching-strategy.md for how to present the choice and record the result in the branching field.
Also confirm: should the JSON file itself be saved on the current branch, or on the task branch? Default to current branch.
10. Write the JSON file
Derive the slug from the PRD filename if available (strip timestamp prefix and extension). If no PRD, derive from the branch name.
Run ~/.dotfiles/claude-code-shared/scripts/scaffolding/task-filename.sh e2e-<slug> to generate the filename.
If a file for this slug already exists (any prefix), ask whether to overwrite or merge (same logic as /to-tasks).
Read the canonical schema now:
cat ~/.dotfiles/claude-code-shared/contracts/task-schema.json
Use that schema exactly. Do not guess field names or structure.
Set "producer": "to-e2e-tasks". Add one e2e-specific top-level field alongside the standard fields:
"source_branch": the branch name these tests were generated from.
HITL tasks (rare — e.g. "provision Playwright auth credentials") must be hands-only: a keyboard action the AI cannot perform. Never emit a decision-review HITL task.
Tell the user the output path and ID range once written.
Output the handoff block:
Next steps:
Then run:
python3 ~/.dotfiles/claude-code-shared/scripts/shared/print-skill-next-steps.py to-e2e-tasks
Append that output under the Next steps header. Do not hardcode skill names.
Read and execute ~/.dotfiles/claude-code-shared/resources/learning-capture.md.
This skill's slug is to-e2e-tasks.
/dispatch-tasks— e2e tasks file is written and ready to implement