feat(engine): add workflow.dir, workflow.file, workflow.name template variables

[PolyphonyRequiem]

Summary

Adds three new template variables available in all agent contexts:

• {{ workflow.dir }} — absolute path to the workflow YAML's parent directory
• {{ workflow.file }} — absolute path to the workflow YAML file itself
• {{ workflow.name }} — workflow name from the YAML config

Motivation

Script agents specify script paths in args, but these resolve relative to CWD (where conductor run is invoked). When workflows live in a registry (e.g., ~/.conductor/registries/twig/), co-located scripts can't be found because CWD is the user's project directory, not the registry.

{{ workflow.dir }} solves this by letting scripts build paths relative to the workflow file:

agents:
  - name: detector
    type: script
    command: pwsh
    args:
      - "-File"
      - "{{ workflow.dir }}/scripts/detect-state.ps1"

This matches how !file tags already resolve prompt file paths relative to the workflow file.

Changes

• src/conductor/engine/context.py: Add workflow_dir, workflow_file, workflow_name fields to WorkflowContext. Include them in build_for_agent() output under the workflow dict (all context modes).
• src/conductor/engine/workflow.py: Set the new fields from workflow_path during WorkflowEngine.__init__().
• tests/test_engine/test_context.py: Tests for metadata in accumulate/explicit modes, and empty-string omission.

Design decisions

• Available in all context modes — workflow metadata is lightweight (3 strings) and fundamental. Unlike agent outputs, these shouldn't be filtered by explicit mode.
• Empty strings omitted — when workflow_path is None (e.g., tests), the fields are empty strings and excluded from the context dict to avoid polluting templates.
• No breaking changes — existing workflow.input behavior is unchanged. New fields are additive.

Testing

• 139 context + workflow engine tests pass
• New tests cover accumulate mode, explicit mode, and empty-metadata omission

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (main@7655dc7). Learn more about missing BASE report.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #121   +/-   ##
=======================================
  Coverage        ?   84.95%           
=======================================
  Files           ?       53           
  Lines           ?     7452           
  Branches        ?        0           
=======================================
  Hits            ?     6331           
  Misses          ?     1121           
  Partials        ?        0           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[jrob5756]

Merge conflict and some small comments. Thanks!

[jrob5756]

Resume drops these fields.

The new workflow_dir/file/name fields aren't included in to_dict() (line 486) or restored by from_dict() (line 502). On resume, cli/run.py:1538 calls engine.set_context(restored_context), which replaces the engine's context (built with the metadata in __init__) with one that has the metadata wiped to empty strings. After resume, {{ workflow.dir/file/name }} will silently disappear from templates — exactly the registry-based script-path scenario this PR exists for.

Recommended fix: repopulate metadata inside WorkflowEngine.set_context() from self.workflow_path / self.config. Keeps absolute paths out of checkpoint files (which would otherwise become stale if the workflow moves) and keeps the source of truth co-located with the engine that knows the path.

Please also add a regression test in tests/test_engine/test_resume.py (or similar) asserting that workflow.dir survives a checkpoint round-trip + resume.

[PolyphonyRequiem]

Yeah, you're right, the round trip drops them and set_context was happily clobbering the engine's view. Fixed in eaf6c9f: set_context now repopulates workflow_dir/file/name from self.workflow_path and self.config after the replace, so the engine stays the source of truth and we keep absolute paths out of the checkpoint. Added test_set_context_repopulates_workflow_metadata that round-trips through to_dict/from_dict, calls set_context, and verifies via build_for_agent that templates actually resolve. Also covered the no-workflow_path case in test_set_context_without_workflow_path_still_sets_name.

[jrob5756]

Suggestion: add an engine-level wiring test.

These tests verify WorkflowContext.build_for_agent() formatting given pre-set fields, but nothing asserts that WorkflowEngine.__init__(config, workflow_path=...) actually wires the path through into those fields (the change at workflow.py:311-315). A small test in tests/test_engine/test_workflow.py like:

def test_engine_populates_workflow_metadata(tmp_path):
    wf_file = tmp_path / "wf.yaml"
    wf_file.write_text("...")  # or use a fixture config
    engine = WorkflowEngine(config, workflow_path=wf_file)
    assert engine.context.workflow_dir == str(tmp_path.resolve())
    assert engine.context.workflow_file == str(wf_file.resolve())
    assert engine.context.workflow_name == config.workflow.name

would catch any future regression in the wiring (e.g., if someone refactors __init__ and reverts to the old WorkflowContext()).

[PolyphonyRequiem]

Fair, there was no guard that __init__ was actually wiring the path into the context. Added test_engine_populates_workflow_metadata and test_engine_workflow_metadata_empty_without_path in tests/test_engine/test_workflow.py to lock that down. Same commit, eaf6c9f.