darnit run has no --framework option although cmd_run reads one

[mlieberman85]

Summary

darnit run has no --framework option, but cmd_run reads one (getattr(args, "framework", None)) and falls back to resolving the framework from repository configuration. Other subcommands (audit, harness, serve) accept -f/--framework. The run subcommand should accept it too, so the framework can be chosen explicitly.

Where

• Parser:

  darnit/packages/darnit/src/darnit/cli.py

  Lines 1199 to 1224 in bbe1bb4

  	run_parser = subparsers.add_parser(
  	"run",
  	help="Run full agentic workflow (LLM-powered)",
  	description="Run the full autonomous compliance pipeline. "
  	"Loads project context, runs all checks, collects context, "
  	"and remediates failures. Requires an LLM API key.",
  	)
  	run_parser.add_argument(
  	"repo_path",
  	nargs="?",
  	default=".",
  	help="Path to repository (default: current directory)",
  	)
  	run_parser.add_argument(
  	"--feedback",
  	dest="feedback_mode",
  	choices=["interactive", "noninteractive", "auto"],
  	default="auto",
  	help="Human feedback mode: interactive (prompts in terminal), "
  	"noninteractive (collects questions for later), "
  	"auto (interactive if terminal, noninteractive in CI)",
  	)
  	run_parser.set_defaults(func=cmd_run)
  	# harness command (feature 026)
  	harness_parser = subparsers.add_parser(

  (the run subparser)
• Consumer:

  darnit/packages/darnit/src/darnit/cli.py

  Lines 665 to 695 in bbe1bb4

  	def cmd_run(args: argparse.Namespace) -> int:
  	"""Run the audit workflow with human feedback.
  	Runs the audit -> collect_context -> remediate pipeline. Checks that
  	require LLM judgement halt for an external agent (e.g. Claude Code);
  	questions needing a human are handled per --feedback mode. Automated
  	in-process LLM backends are not wired into this command yet.
  	"""
  	from darnit.agent.feedback import get_feedback_handler
  	from darnit.agent.graph import audit, collect_context, remediate, route
  	from darnit.agent.state import AuditState
  	repo_path = str(Path(args.repo_path).resolve())
  	# Feedback mode — default to interactive if terminal, noninteractive if not
  	feedback_mode = args.feedback_mode
  	if feedback_mode == "auto":
  	feedback_mode = "interactive" if sys.stdin.isatty() else "noninteractive"
  	print("\nDarnit run")
  	print(f" Repository : {repo_path}")
  	print(f" Feedback : {feedback_mode}")
  	print()
  	# framework_name=None auto-resolves from .baseline.toml inside audit().
  	state = AuditState(
  	local_path=repo_path,
  	framework_name=getattr(args, "framework", None),
  	level=getattr(args, "level", 3),
  	)

  (cmd_run)
• Pattern to copy: the -f/--framework argument on the audit or harness subparsers in the same file.

Getting started

• Add -f/--framework to the run subparser with the same help text style as audit.
• Add a test in tests/darnit/cli/ that darnit run --framework testchecks <fixture> passes the name through (the existing test_cmd_run_e2e.py fixtures are a good starting point).
• Run uv run pytest tests/darnit/cli -q and uv run ruff check .; sign off your commit (git commit -s).

Note: darnit run and darnit audit are development and testing tools (see #505), so this is about making them consistent, not about a user-facing workflow.

[mlieberman85]

Fixed by #556 (darnit run -f/--framework).