feat(workflows): add --dry-run flag to specify workflow run by fuleinist · Pull Request #3124 · github/spec-kit

fuleinist · 2026-06-23T16:25:15Z

Summary

Adds a --dry-run\ flag to \specify workflow run\ that previews step outputs without dispatching AI or shell commands. Per maintainer design guidance (\#2704), the flag lives only on the step-based invocation path — no new CLI commands introduced.

Changes

Engine layer

\StepContext.dry_run\ flag propagated from \WorkflowEngine.execute(dry_run=...)\
\RunState.dry_run\ persisted via save/load, restored on
esume()\
Engine attaches \partial_state\ to exceptions so mid-run failures still surface previews from earlier steps

Step implementations

CommandStep: short-circuits on \context.dry_run, renders \�uild_command_invocation()\ preview when integration is configured, falls back to bare command string
PromptStep: short-circuits with synthetic preview message
GateStep: short-circuits with deterministic first-non-sentinel choice via _coerce_options\ / _first_non_sentinel\ helpers

CLI

\specify workflow run --dry-run\ flag with preview rendering after execution
--json\ suppresses dry-run banner (clean JSON stdout contract)
_print_dry_run_previews()\ shared between \workflow run\ and \workflow resume\

Tests

\ ests/test_dry_run.py: 16 tests covering CommandStep/PromptStep/GateStep dry-run paths, engine execution, resume, and CLI integration

Closes: #2661
Ref: #2704 (closed per maintainer request to split into smaller PRs)

Scoped to engine-only changes per maintainer design guidance: - StepContext.dry_run flag - RunState.dry_run field (persisted via save/load, restored on resume) - WorkflowEngine.execute(..., dry_run=...) + resume() restores it - CommandStep / PromptStep / GateStep short-circuit on context.dry_run with synthetic dry_run_message previews - specify workflow run --dry-run CLI flag + preview rendering - _coerce_options / _first_non_sentinel helpers for deterministic GateStep dry-run branch No new CLI commands introduced. --dry-run lives only on the step-based invocation path (specify workflow run), not on scaffolding commands. Closes: github#2661 Ref: github#2704 (closed per maintainer request, split into smaller PRs)

Copilot

Pull request overview

Adds a --dry-run mode to specify workflow run by propagating a dry-run flag through the workflow engine, persisting it in run state for resume, and teaching key step types (command, prompt, gate) to short-circuit and emit preview output instead of performing side effects.

Changes:

Persist and propagate dry_run across WorkflowEngine.execute(...), RunState.save/load, and resume().
Add dry-run short-circuit behavior to CommandStep, PromptStep, and GateStep.
Add specify workflow run --dry-run CLI flag plus preview rendering, and introduce a dedicated dry-run test suite.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
`tests/test_dry_run.py`	Adds coverage for step-level dry-run behavior, engine persistence/resume, and CLI output expectations (including `--json` behavior).
`src/specify_cli/workflows/steps/prompt/__init__.py`	Implements dry-run short-circuit for prompt steps and standardizes output fields (`dry_run`, `executed`, `dispatched`, etc.).
`src/specify_cli/workflows/steps/gate/__init__.py`	Implements dry-run non-interactive gate resolution with deterministic choice selection and option coercion helpers.
`src/specify_cli/workflows/steps/command/__init__.py`	Implements dry-run preview generation for command invocations without dispatching integration CLIs.
`src/specify_cli/workflows/engine.py`	Adds `dry_run` plumbing, persistence, and `partial_state` attachment for mid-run failures.
`src/specify_cli/workflows/base.py`	Adds `dry_run` to `StepContext` and documents intended preview semantics.
`src/specify_cli/__init__.py`	Adds the `--dry-run` flag, prints previews in non-JSON mode, and adds preview-print helper utilities.

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

mnriem

Please address Copilot feedback

1. GateStep: preserve original output['message'] in dry-run (don't overwrite) 2. CommandStep: preserve original message, set dry_run_message separately 3. PromptStep: same pattern — dry_run_message separate from message 4. --dry-run help text: clarify not all step types short-circuit 5. base.py docstring: accurate about message vs dry_run_message contract 6. _print_dry_run_previews: escape Rich markup in preview text 7. JSON error path: emit valid JSON payload on engine exception 8. resume path: remove stale docstring claim about shared preview helper

Copilot

Copilot's findings

Files reviewed: 7/7 changed files
Comments generated: 9

mnriem · 2026-06-23T19:33:41Z

Please address Copilot feedback

- PromptStep: preserve original prompt in output['message'] - CommandStep: same — keep original command in message - GateStep validate(): accept tuple/list options - _first_non_sentinel: case-insensitive comparison - StepContext docstring: dry_run_message is stable preview surface - Banner: narrow wording to match actual scope - ValueError handler: print partial_state for dry-run - _escape_markup: use rich.markup.escape - workflow_resume: wire dry-run preview output

fuleinist · 2026-06-24T16:11:26Z

Addressed all 9 remaining Copilot review items in c19f6e1:

PromptStep: output["message"] now preserves the original (resolved) prompt instead of the preview string
CommandStep: same — output["message"] keeps the command name
GateStep validate(): now accepts tuple options in addition to list
_first_non_sentinel: case-insensitive comparison (opt.lower() not in ("reject", "abort"))
StepContext docstring: clarifies dry_run_message is the stable preview surface; message preserves original value
Banner: narrowed to "previewing built-in command, prompt, and gate steps without dispatching. Other step types (shell, init) may still execute."
ValueError handler: now prints partial_state dry-run previews (same as generic Exception handler)
_escape_markup: replaced manual [/] escaping with rich.markup.escape
workflow_resume: wired _print_dry_run_previews() for both successful resume and exception handlers when state.dry_run is true

All 15 dry-run tests pass.

Copilot

Copilot's findings

Files reviewed: 7/7 changed files
Comments generated: 3

+from rich.markup import escape as _rich_escape_markup
+
+
+def _escape_markup(text: str) -> str:
+    """Escape Rich markup characters so user-controlled text can be
+    printed safely. Delegates to ``rich.markup.escape`` for canonical
+    handling of ``[``, ``]``, ``{``, ``}``, and other special chars.
+    """
+    return _rich_escape_markup(text)


    except ValueError as exc:
+        if getattr(state, "dry_run", False) and not json_output:
+            _print_dry_run_previews(getattr(exc, "partial_state", None))
        console.print(f"[red]Error:[/red] {exc}")
        raise typer.Exit(1)
    except Exception as exc:
+        if getattr(state, "dry_run", False) and not json_output:
+            _print_dry_run_previews(getattr(exc, "partial_state", None))
        console.print(f"[red]Resume failed:[/red] {exc}")
        raise typer.Exit(1)


+                    try:
+                        preview_invocation = impl.build_command_invocation(
+                            command, args_str
+                        )
+                    except (ImportError, AttributeError, TypeError):


- Move rich.markup.escape import to top-level (fixes E402) - Use exc.partial_state instead of unbound state var in except handlers - Copilot build_command_invocation: return full agent invocation for meaningful dry-run previews in default (non-skills) mode

fuleinist

Addressed all 3 remaining Copilot items from the latest review round:

E402: moved \rom rich.markup import escape\ to top-level imports
UnboundLocalError: except handlers now use \exc.partial_state\ instead of the unbound \state\ variable
Copilot CommandStep preview: \�uild_command_invocation\ now returns \speckit. \ in default mode (was returning just \�rgs), so dry-run previews show the full agent invocation

@mnriem ready for re-review when you have a moment

Copilot

Copilot's findings

Files reviewed: 8/8 changed files
Comments generated: 4

+    console.print("\n[bold yellow]DRY RUN previews:[/bold yellow]")
+    for step_id, result in step_results.items():
+        if not isinstance(result, dict):
+            continue
+        output = result.get("output") or {}
+        if not output.get("dry_run"):
+            continue
+        step_id_display = _escape_markup(str(step_id))
+        preview = output.get("dry_run_message") or output.get("message") or ""
+        preview_escaped = _escape_markup(preview)
+        console.print(f"  [cyan][{step_id_display}][/cyan] {preview_escaped}")


        options = config.get("options", ["approve", "reject"])
-        if not isinstance(options, list) or not options:
+        if not isinstance(options, (list, tuple)) or not options:
            errors.append(
-                f"Gate step {config.get('id', '?')!r}: 'options' must be a non-empty list."
+                f"Gate step {config.get('id', '?')!r}: 'options' must be a non-empty list or tuple."
            )


+        help=(
+            "Preview the workflow without dispatching any AI or shell "
+            "commands for built-in command, prompt, and gate steps. "
+            "Those steps emit a synthetic preview message; the run is "
+            "persisted so it can be inspected but not resumed to a "
+            "real run. Other step types (e.g. init, shell) may still "
+            "perform their normal work during dry-run."
+        ),


    except ValueError as exc:
+        partial = getattr(exc, "partial_state", None)
+        if getattr(partial, "dry_run", False) and not json_output:
+            _print_dry_run_previews(partial)
        console.print(f"[red]Error:[/red] {exc}")
        raise typer.Exit(1)
    except Exception as exc:
+        partial = getattr(exc, "partial_state", None)
+        if getattr(partial, "dry_run", False) and not json_output:
+            _print_dry_run_previews(partial)
        console.print(f"[red]Resume failed:[/red] {exc}")
        raise typer.Exit(1)


mnriem · 2026-06-24T20:37:52Z

Please address Copilot feedback

fuleinist requested a review from mnriem as a code owner June 23, 2026 16:25

Copilot AI review requested due to automatic review settings June 23, 2026 16:25

Copilot started reviewing on behalf of fuleinist June 23, 2026 16:25 View session

Copilot AI reviewed Jun 23, 2026

View reviewed changes

mnriem requested changes Jun 23, 2026

View reviewed changes

mnriem requested a review from Copilot June 23, 2026 19:25

Copilot started reviewing on behalf of mnriem June 23, 2026 19:26 View session

Copilot AI reviewed Jun 23, 2026

View reviewed changes

mnriem requested a review from Copilot June 24, 2026 18:45

Copilot started reviewing on behalf of mnriem June 24, 2026 18:46 View session

Copilot AI reviewed Jun 24, 2026

View reviewed changes

fuleinist commented Jun 24, 2026

View reviewed changes

mnriem requested a review from Copilot June 24, 2026 20:26

Copilot started reviewing on behalf of mnriem June 24, 2026 20:26 View session

Copilot AI reviewed Jun 24, 2026

View reviewed changes

Uh oh!

Conversation

fuleinist commented Jun 23, 2026

Summary

Changes

Engine layer

Step implementations

CLI

Tests

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

mnriem left a comment

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot's findings

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

mnriem commented Jun 23, 2026

Uh oh!

fuleinist commented Jun 24, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot's findings

Uh oh!

fuleinist left a comment

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot's findings

Uh oh!

mnriem commented Jun 24, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants