[Enhancement] Github Workflows: Improve/refactor template for `claude` workflow.

[emcd]

/claude We need to make some modifications to template/.github/workflows/claude.yaml.jinja. Please factor out most of the template into a cross-repo workflow, .github/workflows/xrepo--claude.yaml (with name "Reusable Claude Agent") since we want to be able to use this same workflow logic from multiple repositories but maintain it in only one place. Please look at the other .github/workflows/xrepo--*.yaml workflows for reference and pay particular attention to how they call setup actions, such as python-hatch. The new cross-repo workflow will need to do something similar. You should also look at the various repo-specific workflow templates in template/.github/workflows/*.yaml.jinja to see how they handle workflow versioning and calling the cross-repo workflows.

Please also incorporate a fix that we made to string handling for allowed_tools in .github/workflows/claude.yaml, which was/is our prototype for the workflow template. The template's current handling of allowed tools may be broken, since the arguments must be space-separated.

And, please consider whether the Jinja {% raw %} directives are necessary around the trigger conditions for the claude job. (They may be, but it is not obvious to me that they are. Please include a justification in your final report or PR, if they are, or else remove them.) They are obviously necessary around any ${{ }} to suppress Jinja processing and preserve the structure for Github Actions, but I am less certain about other cases.

[emcd]

/claude Please work on this issue. Also, note that to validate the changes, you should run hatch run emcdproj template validate --preserve default. (You may need to do pip install copier first.) This will use Copier to create a project from the template in a temporary directory. The --preserve flag will prevent the temporary directory from being cleaned up and this will allow you to inspect that the resulting <tmp dir>/.github/workflows/claude.yaml looks correct. (You will be able to infer the temp directory's location from the command output.)

[claude]

Claude finished @emcd's task —— View job • Create PR ➔

---

Task: Refactor Claude Workflow Template

Completed issue #8 - Created a reusable cross-repo workflow for the Claude agent.

Todo List:

• Read repository guidelines (CLAUDE.md)
• Examine current Claude workflow files
• Study existing xrepo workflows for patterns
• Analyze repo-specific templates for versioning
• Create xrepo--claude.yaml workflow
• Update claude.yaml.jinja template
• Fix allowed_tools string handling
• Evaluate {% raw %} directive necessity
• Complete refactoring and document changes

Changes Made:

1. Created .github/workflows/xrepo--claude.yaml:

   • Reusable workflow following the established xrepo pattern
   • Configurable inputs: python-version, timeout-minutes, allowed-tools
   • Proper secrets interface for anthropic-api-key
   • Uses emcd/python-project-common/.github/actions/python-hatch@master for setup

2. Updated template/.github/workflows/claude.yaml.jinja:

   • Now calls the xrepo workflow with @{{ _commit }} versioning
   • Fixed allowed_tools from pipe-separated to space-separated format using YAML >- folding
   • Removed unnecessary {% raw %} directives around trigger conditions
   • Maintains all existing functionality while improving maintainability

Analysis Results:

{% raw %} Directive Evaluation: The {% raw %} directives around trigger conditions were unnecessary because:

• They were wrapping plain GitHub Actions expressions like github.event_name == 'issue_comment'
• {% raw %} is only needed around ${{ }} expressions to prevent Jinja processing
• The trigger conditions don't contain ${{ }} syntax, so Jinja processing isn't a concern

allowed_tools Fix: Changed from pipe-separated format to space-separated format matching the working .github/workflows/claude.yaml implementation.

The refactoring successfully factors out the workflow logic into a reusable component that can be maintained in one place and used across multiple repositories, while preserving all functionality and improving consistency.

Note: Files are ready for commit. Cannot push changes due to permission limitations, but all changes are locally complete and validated.