feat(workflows): allow resume to accept workflow inputs

[doquanghuy]

Description

Closes #2812.

workflow run accepts --input key=value, but a paused/failed run could only be resumed with the inputs captured at start time. This lets workflow resume accept the same --input flag, so a run can continue with information that became available after it paused (or with a corrected value after a failure).

specify workflow run wf.yml          # pauses/fails needing a value
specify workflow resume <run_id> --input cmd="exit 0"   # resumes with it

What changed

• workflow resume gains --input/-i, parsed by a new shared _parse_input_values() helper now used by both run and resume (no duplicated parsing).
• WorkflowEngine.resume(run_id, inputs=None): supplied values are merged over the run's persisted inputs and re-resolved through the existing _resolve_inputs typed-validation path, then persisted. The resumed step (re-run from the pause/failure point) sees the updated inputs.
• Reference docs updated (docs/reference/workflows.md).

Compatibility

• Reuses the existing input model and typing — not a new answer/gate-specific mechanism (deliberately avoids the feat(workflows): add --gate-script for non-interactive gate testing #2667 direction).
• Resuming without --input is unchanged; keys not supplied keep their persisted values; ill-typed values fail fast with the same error as run; undeclared keys are ignored exactly as run does today.
• Distinct from [Feature]: Support file references for workflow inputs #2405 (file-reference inputs at run time); this is resume-time input continuation.

Testing

• Tested locally with uv run specify --help
• uv sync --extra test && uv run pytest — 3312 passed, 40 skipped; ruff check src/ clean
• Manually verified: a run that fails on {{ inputs.cmd }}="exit 1" completes after resume --input cmd="exit 0", with inputs.json updated; invalid --input format rejected on both run and resume

Tests in tests/test_workflows.py::TestResumeWithInputs cover: resume re-runs the step with the new value; resume without input preserves inputs; typed coercion + persistence on resume; invalid type raises; CLI --input invalid-format error.

AI Disclosure

• I did not use AI assistance for this contribution
• I did use AI assistance (describe below)

Used Claude to implement the flag/engine change, tests, and docs. Behaviour was verified locally and the diff reviewed before submission.

---

Per CONTRIBUTING.md (new CLI arguments should be agreed first), #2812 is the discussion issue and this is the concrete proposal — happy to adjust or hold. It touches workflow resume like #2814 does; trivial to rebase whichever lands first. @mnriem a quick steer on whether you want resume-time inputs would be appreciated.

[Copilot]

Pull request overview

This PR extends the Spec Kit workflow system to support supplying updated workflow inputs at resume time, reusing the same typed input validation/coercion path as initial runs. This aligns specify workflow resume with specify workflow run by allowing --input key=value overrides to be merged into the persisted run inputs and then persisted back to the run state.

Changes:

• Add --input/-i to specify workflow resume, sharing parsing logic with workflow run via a new _parse_input_values() helper.
• Extend WorkflowEngine.resume(run_id, inputs=None) to merge provided inputs over persisted inputs and re-run _resolve_inputs() for typed coercion/validation, persisting the updated inputs.
• Add tests covering resume-time input override/merge/coercion/persistence and CLI invalid-format errors, plus reference doc updates.

Show a summary per file

File	Description
src/specify_cli/__init__.py	Adds shared CLI --input parsing helper and wires --input/-i into workflow resume.
src/specify_cli/workflows/engine.py	Updates WorkflowEngine.resume() to accept/merge inputs and re-validate/coerce them via existing _resolve_inputs.
tests/test_workflows.py	Adds coverage for resume-time input overrides (engine + CLI) including persistence and type validation.
docs/reference/workflows.md	Documents workflow resume --input behavior and option table.

Copilot's findings

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

• Files reviewed: 4/4 changed files
• Comments generated: 0

[doquanghuy]

@mnriem this one's ready for review. I also audited it against the two issues Copilot raised on the sibling PRs — non-string coercion of an evaluate_expression result (#2810) and a doc example not matching real CLI output (#2814). Neither applies here: this PR doesn't evaluate any templates, and the docs show only a command invocation (no sample output), so no changes were needed.