Taskfile.yaml

# https://taskfile.dev
---
version: "3"

vars:
  TEMPLATE_REVISION:
    sh: git rev-parse HEAD

tasks:

  default:
    summary: |
      Just a list of documented tasks.
    silent: true
    cmds:
      - task --list

  clean:
    desc: clean up working directory
    cmds:
      - rm -rf *_dir

  create:
    desc: create a project with the current version and ask for answers
    deps:
      - clean
    cmds:
      - copier copy --trust --vcs-ref {{.TEMPLATE_REVISION}} . new_dir

  check:generate:case:
    desc: generate a project with the current version from a TEST_CASE
    preconditions:
      - sh: "[[ -v TEST_CASE ]]"
        msg: >
          Environment needs 'TEST_CASE' variable.
          See tests directory - all yaml files without suffix are test cases.
      - sh: "[ -f tests/${TEST_CASE}.yml ]"
        msg: Test case file ${TEST_CASE}.yml does not exist.
    cmds:
      - rm -rf {{.TEST_CASE}}_dir
      - mkdir {{.TEST_CASE}}_dir
      - cp tests/{{.TEST_CASE}}.yml {{.TEST_CASE}}_dir/.copier-answers.yml
      - copier copy --trust --vcs-ref {{.TEMPLATE_REVISION}} --defaults . {{.TEST_CASE}}_dir
      - cd {{.TEST_CASE}}_dir && git init -b develop .
      - cd {{.TEST_CASE}}_dir && git add .
      - cd {{.TEST_CASE}}_dir && git config user.name "Anonymous Person"
      - cd {{.TEST_CASE}}_dir && git config user.email "anonymous@example.org"
      - cd {{.TEST_CASE}}_dir && git commit -m "init from template"

  check:generate:cases:
    desc: generate all test cases with the current template version
    deps:
      - clean
    vars:
      TEST_CASES:
        sh: find tests -type f -name '*.yml' | sed 's/^tests\///g' | sed 's/\.yml$//g'
    cmds:
      - for: {var: TEST_CASES}
        cmd: TEST_CASE={{.ITEM}} task check:generate:case

  check:hook:case:
    desc: smoke test the Stop hook a generated package ships
    summary: |
      Nothing else here ever executes the shipped agent files, and this one is
      a blocking Stop hook: if it stayed silent when a template owned file was
      edited the feature would be dead, and if it spoke on a clean tree or
      ignored the loop guard every session in every generated package would
      stop being able to end. All three are asserted.

      This cannot tell you the hook fires on the right evidence in a real
      repository. That is still a hand run in a rendered case - see CLAUDE.md.
    cmds:
      - cd {{.TEST_CASE}}_dir && ../scripts/check-hook.py

  check:skillhook:case:
    desc: smoke test the skill check hook a generated package ships
    summary: |
      The hook runs on every edit and shell command in every generated package,
      so one that decided on a tool call would stall sessions, and a mute one
      would remind of nothing. Both directions are asserted: an edit of a
      mapped file, and a command naming one, get a reminder naming the skill;
      unmapped files, prose and code get none; the output is never more than
      context; and a user-level copy reads its own map for a package without
      one and stands down where the package's own copy runs.
    cmds:
      - cd {{.TEST_CASE}}_dir && ../scripts/check-skill-hook.py

  check:skills:case:
    desc: assert the skills a generated package ships are present and well formed
    summary: |
      Nothing here can test what a skill says - that is a hand read. What this
      asserts is that they ship: frontmatter parses, the name matches the
      directory, no Jinja leaked into a rendered file, and the two skills that
      are conditional on package_type are present for a project package and
      absent for a vocabulary one.
    cmds:
      - cd {{.TEST_CASE}}_dir && ../scripts/check-skills.py

  check:offline:case:
    desc: assert the offline RDF checkers a generated package ships still bite
    summary: |
      A generated package ships one example.ttl that declares no shapes and no
      vocabulary terms, so every check is a no-op over it and a checker that
      had stopped detecting anything would look exactly as healthy as one that
      works. This drives the rendered package's own bin/ over tests/fixtures
      instead - a good catalog that must pass, and one planted fault per
      checker that each must catch by name.

      It needs no Corporate Memory connection, so it is the part of
      check:validate:case that still runs with no credentials.
    cmds:
      - cd {{.TEST_CASE}}_dir && ../scripts/check-offline.py

  check:validate:case:
    desc: validate a project test case
    preconditions:
      - sh: "[[ -v TEST_CASE ]]"
        msg: >
          Environment needs 'TEST_CASE' variable.
          See tests directory - all yaml files without suffix are test cases.
      - sh: "[ -d ${TEST_CASE}_dir ]"
        msg: >
          Test case directory ${TEST_CASE}_dir does not exist.
          Please use check:generate* tasks to create directories.
    cmds:
      - task: check:hook:case
        vars: {TEST_CASE: '{{.TEST_CASE}}'}
      - task: check:skillhook:case
        vars: {TEST_CASE: '{{.TEST_CASE}}'}
      - task: check:skills:case
        vars: {TEST_CASE: '{{.TEST_CASE}}'}
      - task: check:offline:case
        vars: {TEST_CASE: '{{.TEST_CASE}}'}
      - cd {{.TEST_CASE}}_dir && task check

  check:validate:cases:
    desc: validate all test cases
    vars:
      TEST_CASES:
        sh: find tests -type f -name '*.yml' | sed 's/^tests\///g' | sed 's/\.yml$//g'
    cmds:
      - for: {var: TEST_CASES}
        cmd: TEST_CASE={{.ITEM}} task check:validate:case

  check:
    desc: First generate, the validate all test cases
    cmds:
      - task: check:generate:cases
      - task: check:validate:cases