feedback_inbox and app_errors guessed kuhyx/<folder> for a task's repo, which filed cleanup-plate under kuhyx/cleanup-plate (real: Sepci0/cleanup-plate) and showed phantom repos on the board. Both now read the GitHub owner/name from ~/src/<folder>'s origin remote (https or ssh forms) and fall back to the guess only when there is no checkout or its origin is not on GitHub. app-errors 0.1.2. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WzxdWZfgt44aEKxvGBJkwP |
||
|---|---|---|
| .. | ||
| feedback_inbox | ||
| systemd | ||
| .pre-commit-config.yaml | ||
| install.sh | ||
| pyproject.toml | ||
| README.md | ||
| requirements.txt | ||
feedback_inbox
Turns every game's dev-feedback bundle into one todo-app task, so feedback kuhy saves with F12 is never forgotten.
A game saves ~/data/<repo>_binaries/feedback/<stamp>/ (report.json,
screenshots, logs) and appends a ready-to-run Claude prompt to
feedback-<date>.md beside it. This package imports each complete bundle once
as a task in the todo app for the GitHub repo named by the origin of
~/src/<repo> (kuhyx/<repo> only when there is no such checkout or it is not
on GitHub).
Task shape
- Title:
MANUAL WORK REQUIRED: <repo> feedback: <first line of his note>(note capped at 120 characters). The prefix makes todoImplementer list it as a human part and never run it. No note:... feedback: <game> <combo> <HH:MM>. Areport.jsonthat does not parse still gets a task, tagged[corrupt report.json]; the note then comes from the markdown entry. - Body: a
Due: <next day 10:00>line, then the bundle's markdown entry (the prompt with absolute paths). - The task is its own note (
add_taskwithnew_note). The server stores that as<repo> todos: <title>plus a## <title>section, so right after adding, the inbox rewrites it withupdate_taskto the exact title and a section-free body (the entry's own##headings become###). todoImplementer splits notes at##headings, and a title that does not start withMANUAL WORK REQUIRED:would make a runnable part. The marker recordsformat: 2once the rewrite worked; a failed rewrite is retried on the next sweep (never a second task), and markers without it are repaired the same way. - Closing the task cancels the feedback. "Act on feedback" in a Claude session
means working every open
<repo> feedback:task.
The todo app has no due-date or reminder field yet, so the due time is only text in the body.
How it runs
| Unit | Does |
|---|---|
feedback-inbox.path |
PathChanged on every existing *_binaries/feedback folder; runs feedback-inbox run |
feedback-inbox-sync.path |
PathChanged on ~/data and every *_binaries folder; runs sync then run |
feedback-inbox-sync.timer |
every 5 min, the same sweep (events during a run are lost by .path units) |
.path units take no globs: feedback-inbox sync writes the folder list into
the drop-ins feedback-inbox{,-sync}.path.d/dirs.conf and reloads systemd only
when the set of games changed.
A bundle is complete when report.json is at least 2 s old and its markdown
entry exists (matched by the stamp folder in the entry's paths). Incomplete
bundles are retried; run waits up to 15 s for them.
The import marker is .inbox-imported.json inside the bundle (task id, title,
due). It is written only after the todo server accepted the task, so a server
outage leaves no marker, exits non-zero, and the next sweep retries.
Install
scripts/run_subproject_tests.sh feedback_inbox # from the monorepo root
feedback_inbox/install.sh
install.sh builds a stdlib-only venv, links feedback-inbox into
~/.local/bin, mints a todo-server token into ~/.config/feedback-inbox/token
(0600; revoke by deleting its line in ~/.config/todo-mcp/tokens) and enables
the units. Enabling sweeps once, so existing bundles are imported then.
Tests
pytest from this directory (100 % branch coverage enforced). Tests use a temp
HOME, a fake todo client, and a local HTTP server for the MCP client; the
autouse conftest.py redirects the one path resolver and points the todo URL
at a dead port.