# fix-and-test — reproduce a bug with a failing test, then fix it.
#
# Two agent steps rather than one, on purpose. Asking a single agent to "fix
# the bug and add a test" reliably produces a test written to pass against the
# fix it just made. Splitting the work makes the first step's output — a test
# that *fails* — a thing the second step has to satisfy.
#
# The first step's check is inverted: it succeeds when the test suite fails,
# because at that point a red test is the deliverable.
name: fix-and-test
description: Write a failing test that reproduces a bug, then make it pass.

defaults:
  agent: claude
  max_retries: 2
  timeout: 30m

steps:
  - id: reproduce
    type: agent
    prompt: |
      Reproduce this bug with a failing test. Do not fix it yet.

      Bug: {{.Task.Title}}
      {{with .Task.Description}}
      Details:
      {{.}}
      {{end}}

      Add the smallest test that fails because of this bug, in the package
      where the bug lives, following the testing conventions already in this
      repository. Change nothing else. When you are done the test suite must
      fail, and it must fail on your new test.
    # Inverted deliberately: this step is done when the suite is red.
    # `!` is understood by bash, cmd and PowerShell 7 alike (§8.3).
    check: '! go test ./...'
    check_timeout: 15m

  - id: fix
    type: agent
    prompt: |
      A previous step added a failing test that reproduces this bug:

      {{.Steps.reproduce.Result}}

      Now make it pass. Fix the underlying cause rather than the symptom, and
      do not weaken, skip or delete the test. Leave the rest of the suite
      green.
    check: go test ./...
    check_timeout: 15m

  - id: commit
    type: command
    run: 'git add -A && git commit -m "fix: {{.Task.Title}}"'

  - id: review
    type: manual
    instructions: |
      Check that the test in task #{{.Task.ID}} genuinely fails without the
      fix — `git stash` the non-test changes and run it — and that the fix
      addresses the cause rather than the assertion.