fix(workflows): reject a non-string 'integration'/'model' in command & prompt steps

[Noor-ul-ain001]

Problem

A command or prompt step whose integration: (or model:) is a non-string crashes the run with a raw TypeError instead of a clean validation/step error.

In execute(), the resolved integration value is passed to get_integration(), which does INTEGRATION_REGISTRY.get(key) — using the value as a dict key. An unhashable value (list/dict) raises TypeError: unhashable type right there:

>>> from specify_cli.workflows.steps.command import CommandStep
>>> from specify_cli.workflows.base import StepContext
>>> CommandStep().execute({"id": "x", "command": "speckit.specify", "integration": ["claude"]}, StepContext())
TypeError: cannot use 'list' as a dict key (unhashable type: 'list')

Crucially this is not limited to unvalidated runs: neither the step's validate() nor validate_workflow checked the type of integration/model, so validate() returns [] for integration: [claude] and the crash survives validation. A non-string model similarly reaches build_exec_args() and would be fed into the CLI argv.

This is the same recurring shape as the existing command/prompt/input/options guards — a string-typed dispatch field that execute() passes downstream without a type check.

Fix

Both command and prompt steps:

• validate() — reject a literal non-string integration/model, mirroring the existing sibling type checks. An explicit YAML-null (integration: → inherit the workflow default) and a "{{ ... }}" expression both stay valid.
• execute() — guard the resolved value (after the or-fallback to the workflow default and expression evaluation) and return a clean FAILED StepResult rather than letting the run crash. This also catches an expression that resolves to a non-string and an unvalidated workflow-level default.

Tests

Added parametrized validate() rejection tests, "accepts None / expression" tests, and execute()-fails-loudly tests to both TestCommandStep and TestPromptStep. Test-the-test: with the fix stashed, the new tests fail — including the raw TypeError: unhashable type traceback at get_integration. Full tests/test_workflows.py shows no new failures (the 20 pre-existing failures are all Windows symlink-guard tests, identical on a clean tree).

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Adds validation and runtime guards for non-string command/prompt integration and model values.

Changes:

• Rejects invalid literal values during validation.
• Returns failed step results for invalid resolved values.
• Adds validation and execution tests.

Show a summary per file

File	Description
src/specify_cli/workflows/steps/command/__init__.py	Validates command dispatch fields.
src/specify_cli/workflows/steps/prompt/__init__.py	Validates prompt dispatch fields.
tests/test_workflows.py	Tests new validation and failure behavior.

Review details

Tip

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

• Files reviewed: 3/3 changed files
• Comments generated: 2
• Review effort level: Medium

[mnriem]

Please address Copilot feedback and resolve conflicts

[Noor-ul-ain001]

@
Addressed the Copilot feedback and resolved the merge conflicts.

Copilot feedback (falsey non-string bypass). Both flagged spots resolved the value with config.get("integration") or context.default_integration (and the model equivalent), which coerces a falsey non-string — [], {}, 0, False — into the workflow default before the new type guard runs. Since execute() is explicitly supported without prior validation, such a value was silently accepted and, with a configured default, could dispatch using the wrong integration/model instead of returning the contract error.

Fixed in both command and prompt steps by falling back to the workflow default only for a genuinely-unset value (missing / YAML-null / empty string):

integration = config.get("integration")
if integration is None or integration == "":
    integration = context.default_integration

so every non-string now reaches the guard.

Tests. Added parametrized execute() cases over [], {}, 0, False to both TestCommandStep and TestPromptStep, each with default_integration/default_model set so a regression would dispatch-with-the-default rather than fail-not-possible. Test-the-test: with the source fix stashed, all 8 fail; with the fix, they pass.

Conflicts. Merged upstream/main; the only conflict was in tests/test_workflows.py where both sides added new test methods to TestCommandStep (the incoming non-string command guard tests and these integration/model tests) — kept both.

Full tests/test_workflows.py: 677 passed, 7 skipped, 20 failed — the 20 are all the pre-existing Windows symlink-guard tests, identical on a clean tree and unrelated to this change.
@