[zwanski-start]

Closes #11

Problem

Features the agent can't verify on its own (the reporter's case is a native iOS app with no simulator available to the agent) never reach Done. Users had no way to close them from the UI after checking them by hand.

Changes

• Backend (server/routers/features.py): two new endpoints:
  • POST /api/projects/{name}/features/{id}/mark-passing sets passes=True and clears in_progress and needs_human_input.
  • POST /api/projects/{name}/features/{id}/reopen sets passes=False and clears the same flags, which puts the feature back in Pending so agents pick it up again.
• UI:
  • FeatureModal now has a Mark as Passing button next to Edit/Skip on any feature that isn't done.
  • Done features now have a footer with a Reopen button. Before this, done features had no actions at all.
  • Added markFeaturePassing/reopenFeature in lib/api.ts and useMarkFeaturePassing/useReopenFeature hooks. These refresh the features query when they succeed.

Verification

• ruff check, npm run lint (0 errors, only warnings that were already there) and npm run build all pass.
• Ran the backend with an isolated HOME and seeded a project with 2 features:
  • Called both endpoints with curl. The feature moved Pending → Done → Pending as expected, and an unknown ID returns 404.
  • In the browser (Playwright): opened the feature in Kanban, clicked Mark as Passing, and the card moved to Done. Reopened it, clicked Reopen, and it went back to Pending.

Assumptions / notes

• I didn't add a separate "Code Complete / Testing" lane. A human can now close any card from the existing lanes, and the conditional Needs Input lane already exists for features that are waiting on a person.
• Manual marking is allowed even while a feature is in progress. If the agent marks it passing again later, nothing breaks.
• Follow-up worth doing: feature_request_human_input exists in the MCP server but isn't in CODING_AGENT_TOOLS in client.py. That means coding agents can't currently move a feature they can't test into the Needs Input lane. I left the agent tool permissions alone to keep this PR focused.

🤖 Generated with Claude Code

[zwanski-start]

🔍 QA report: ✅ Passed

Tester: Barbara (cubefarm agent, in a fresh session) · Round: 1 · Author: Ada

The PR adds working POST /mark-passing and /reopen endpoints and matching FeatureModal actions, which meets the issue's request for a manual 'mark as passed' button for features the agent can't test. I verified this via curl, including the in-progress, needs-input and error cases, and in a real browser: a card moved Pending → Done → Pending and the live progress updated. Lint, type checks, build and pytest match master with no new failures. One non-blocking issue: at phone width (375px) the extra footer button pushes the Delete icon outside the modal, so it can only be reached by scrolling the dialog sideways.

	Check	Details
✅	Issue acceptance: user can manually mark an untestable feature as passing	Opened a Pending feature in Kanban and clicked 'Mark as Passing'. The modal closed, the card moved to Done, and the Dashboard progress went from 1/4 to 2/4 within the 2s poll.
✅	Reopen a done feature	Done features now show a Reopen footer. Clicking it returned 'Login screen' to Pending and the Done count dropped from 2 to 1.
✅	API: mark-passing clears in_progress and needs_human_input	On seeded feature #2 (in_progress) and #3 (needs_human_input with a request), mark-passing returned passes=true, in_progress=false, needs_human_input=false. GET /features listed both in done.
✅	API: reopen is idempotent and returns the feature to pending	Reopening #4 twice returned passes=false both times with no error.
✅	API error and validation cases	Unknown feature ID 999 → 404 on both endpoints. Unknown project → 404. Non-numeric ID → 422. Invalid project name → 400. GET on the POST route → 404.
✅	Dependency graph reflects manual status	The Graph view rendered #1 and #4 as Done (green) and #2 and #3 as Pending, with the #1→#4 edge.
✅	Browser console	0 errors. 4 warnings, all Radix 'Missing Description' for DialogContent, which FeatureModal already produced on master.
❌	Responsive layout at 375px	Non-blocking. With 'Mark as Passing' added, the footer content is 405px wide in a 343px dialog. The Delete icon (x=370–406) sits outside the dialog edge (359), the 'Priority' label is clipped, and the dialog scrolls sideways (overflow-x:auto). Removing the new button in DevTools restores the fit (Delete at x=342), so the PR causes it. Delete is still reachable by scrolling, so nothing is broken.
⏭️	Sidebar project % badge refresh	The sidebar badge stays at 25% until reload because the hooks invalidate only ['features']. Existing skip/delete hooks behave the same way, so it isn't a regression from this PR.
✅	Security / schema / leftover code	No changes to security.py, the localhost middleware or the DB schema. No debug code or logging of sensitive data. Errors are logged with logger.exception and return a generic 500.
⏭️	New logic covered by tests	No backend tests were added for the new endpoints. The repo has no existing router tests to extend. I covered the behaviour manually instead.
⏭️	UI Playwright e2e suite	Could not run: the Chromium headless shell binary is missing and npx playwright install chromium timed out after 300s. The existing specs (assistant panel, tooltips) don't touch FeatureModal.

🧪 Commands run

Command	Result
git diff origin/master...HEAD	4 files, +155/-5 (features router, FeatureModal, useProjects, api.ts)
ruff check .	All checks passed
mypy . (PR vs origin/master worktree)	Same errors as master (25 vs 24; the extra one is in venv/bin/activate_this.py). None in server/routers/features.py
python -m pytest -q (PR vs origin/master)	144 passed, 21 failed. The same 21 failures (test_devserver_security etc.) on master
npm --prefix ui run build	tsc -b and vite build succeeded
npm --prefix ui run lint	0 errors, 6 warnings (already present)
HOME=<isolated> uvicorn server.main:app --host 127.0.0.1 --port 5443	Booted; /api/health returned 200
curl -XPOST .../features/{2,3}/mark-passing, .../4/reopen x2, error cases	200 with correct flags; 404/404/422/400 for the invalid cases
npx playwright test --reporter=line (ui)	15 failed: browser executable missing (setup problem). npx playwright install chromium timed out

No browser screenshots were taken in this round.

_{Posted by cubefarm · merges automatically once GitHub's checks pass}