cnb: dispatcher keep lead working, not nudge dev idle (#223) - #244
Merged
ApolloZhangOnGithub merged 6 commits intoMay 17, 2026
Merged
Conversation
Port the bash version-check from `bin/cnb` into a reusable Python module so tongxue who run `board` directly (skipping the `cnb` wrapper) also detect a stale install and route an update task to the device-supervisor tongxue. Each tongxue does not self-update. - `lib/update_check.py`: pure-Python check_update + helpers, shares the existing on-disk cache (`~/.cnb/latest-version`) and notification suppression (`~/.cnb/update-notified`) with the bash version. - `bin/board`: silent startup hook in main(); never blocks dispatch. `CNB_SKIP_UPDATE_CHECK=1` disables for tests / quick runs. - `board update-check [--force]`: manual trigger for debugging. - Skipped in venv (user-managed install). - Cross-provider versions are normalized: PEP 440 .dev0, npm -dev, v-prefix, prerelease suffixes all compare correctly. - Steady-state overhead is sub-millisecond: cache hit + tuple compare; npm is only re-fetched in the background when the 60-min TTL expires. - 36 unit tests cover normalization, venv detection, owner resolution precedence, notification suppression, async refresh, and the CLI command. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Follow-up to PR #224. Removes the duplicate version-check logic from bin/cnb (~120 lines) so both entry points (`cnb <subcmd>` and the interactive banner) share one implementation in lib/update_check.py. - `bin/cnb` subcommand path (line 165): replaced `_check_update notify` with `bin/board update-check --quiet` (silent stdout; owner still gets notified when stale). - `bin/cnb` banner path: replaced `_check_update` with `bin/board update-check --terminal` (silent unless stale, then prints the historical yellow banner line so UX is unchanged). - Removed `_check_update`, `_notify_update_owner`, `_version_gt`, `_in_virtualenv`, `_CACHE`, `_NOTIFIED` from bash. Kept `_read_update_owner` because `cnb exec` still uses it for sender fallback. - Added `--quiet` and `--terminal` modes to `cmd_update_check`, with tests for each (silent up-to-date, silent stale-quiet, yellow banner on stale-terminal, no banner in venv). - `bin/board` startup hook now skips when the command itself is `update-check` to avoid double-firing the check. 41 update-check tests pass; full suite 1670/1670 (excluding the pre-existing test_board_msg drift bezos is fixing in PR #220). Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Both from the PR #224 review: 1. Hoist the `is_venv()` check to the top of `cmd_update_check`. The old order ran `--force`'s suppression-clear and 2s sleep before `check_update`'s internal venv guard kicked in, so debugging from a venv shell wasted an npm spawn + wait. 2. Replace `time.sleep(2)` after the async refresh with a new `refresh_latest_version_sync(timeout=10)` helper. The old sleep was a race: a slow npm finished after the sleep returned, defeating `--force`; a fast one made the user wait for nothing. The sync variant blocks until npm returns or times out. Net: 6 new tests (sync-refresh success / timeout / missing npm / non-zero exit / empty stdout; venv-short-circuits-before-refresh). 42/42 update_check tests pass. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The subcommand-path hook was synchronously waiting for `bin/board update-check --quiet` to complete before dispatching, which surfaced as an empty-stdout failure on Linux CI for `cnb projects scan --json` (local pass, CI fail across 3.11/3.12/3.13). Detach the hook with `( ... ) & disown` plus full fd redirect so it cannot block or pollute the downstream subcommand whose stdout the test captures. Local repro of the test still passes; this is a defensive bet for the Linux CI failure. The banner path (line ~486) keeps its synchronous `--terminal` call since the user is sitting at the terminal anyway. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reverts the bin/cnb subcommand-path consolidation from #228. Replacing the inline bash `_check_update notify` with `bin/board update-check --quiet` caused two failures on Linux CI that don't repro locally: 1. `test_bin_cnb_projects_scan_dispatches_json_contract` — empty stdout from `cnb projects scan --json` (sync hook somehow interferes with downstream subcommand output capture on Linux). 2. `test_version_subcommand_notifies_lead_when_outdated` — when the prior commit detached the hook with `& disown` to avoid #1, the notification stopped being delivered before the test's inbox check. Either path (sync / async) breaks a different test. The bash version was working before; restore it and let the bin/board startup hook remain as the deliverable for #43. Both paths are now in place — a small amount of duplication, but it's safer to ship than to keep fighting CI. The bin/board hook (the actual KR2 value — tongxue running `board` directly get the check) is unchanged. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Inverts the previous nudge model: - Workers (dev) idle is normal — they wait for lead to assign work. Remove the "OKR continue" nudge that interrupted them with self-driven work prompts. - Lead idle is anomalous — the org has no engine. Add explicit lead keep-alive: when lead is idle, nudge it to scan team status, review PR queue, and proactively dispatch the next issue to free workers. NudgeCoordinator.tick now processes lead separately with a different message and only inbox/lead_idle nudge types (no queued_flush for lead since it's the org root). Bump VERSION to 0.5.79-dev. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
4 tasks
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 573edcf8cc
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if identity: | ||
| validate_identity(db, identity) | ||
|
|
||
| _maybe_check_update(env, cmd.name) |
There was a problem hiding this comment.
When a stale cache exists and no update-notified key has been written yet, this hook also runs for board --as dispatcher send ...; the update notifier itself sends its message by spawning bin/board --as dispatcher send, so the child process re-enters _maybe_check_update before dispatching the send and recursively spawns more board sends until timeouts rather than delivering the notification. This affects both the new Python notifier and the existing bin/cnb bash notifier whenever they try to send the first stale-version message; skip the hook for send or set CNB_SKIP_UPDATE_CHECK=1 on the internal send.
Useful? React with 👍 / 👎.
| @@ -1 +1 @@ | |||
| 0.5.76-dev | |||
| 0.5.79-dev | |||
There was a problem hiding this comment.
VERSION is the canonical input for bin/sync-version --check, which expects package.json to match it and pyproject.toml to contain the PEP 440 form. This change sets VERSION to 0.5.79-dev while the other two files are 0.5.78, so the documented PR/release check fails and the installed runtime version can diverge from the package metadata.
Useful? React with 👍 / 👎.
5 tasks
ApolloZhangOnGithub
added a commit
that referenced
this pull request
May 17, 2026
Three failures introduced by #244 (lead-keepalive refactor) that broke master CI for every downstream PR: 1. **typecheck**: lib/concerns/nudge_coordinator.py:184 had mixed lambda signatures in a list — one with `n="lead"` default, one bare. mypy couldn't infer a unified callable type. Normalized both to no-arg lambdas; the "lead" string is now closed over directly. 2. **lint**: lib/concerns/helpers.py had `# noqa: F401 — re-export for concerns` on `tmux_ok` that ruff RUF100 flagged as unused (downstream usage made F401 already silent). Removed redundant directive. 3. **check-consistency**: #244 bumped VERSION to 0.5.79-dev without running `bin/sync-version`, leaving pyproject.toml and package.json at 0.5.78-dev. Synced all three, then bumped to 0.5.80-dev for this hotfix. Verified locally: - ruff check on full CI invocation: clean - mypy lib/: 65 source files, no issues - bin/sync-version --check: OK - bin/check-{changelog,branding,readme-sync,site-docs}: all OK Same shape as PR #222 (round 1) and PR #230 (round 2) — operational hotfix per CLAUDE.md rule 5 ("operational issues get assigned and executed, not discussed"). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This was referenced May 17, 2026
ApolloZhangOnGithub
added a commit
that referenced
this pull request
May 17, 2026
Three failures introduced by #244 (lead-keepalive refactor) that broke master CI for every downstream PR: 1. **typecheck**: lib/concerns/nudge_coordinator.py:184 had mixed lambda signatures in a list — one with `n="lead"` default, one bare. mypy couldn't infer a unified callable type. Normalized both to no-arg lambdas; the "lead" string is now closed over directly. 2. **lint**: lib/concerns/helpers.py had `# noqa: F401 — re-export for concerns` on `tmux_ok` that ruff RUF100 flagged as unused (downstream usage made F401 already silent). Removed redundant directive. 3. **check-consistency**: #244 bumped VERSION to 0.5.79-dev without running `bin/sync-version`, leaving pyproject.toml and package.json at 0.5.78-dev. Synced all three, then bumped to 0.5.80-dev for this hotfix. Verified locally: - ruff check on full CI invocation: clean - mypy lib/: 65 source files, no issues - bin/sync-version --check: OK - bin/check-{changelog,branding,readme-sync,site-docs}: all OK Same shape as PR #222 (round 1) and PR #230 (round 2) — operational hotfix per CLAUDE.md rule 5 ("operational issues get assigned and executed, not discussed"). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Inverts the previous nudge model:
Removed the "OKR continue" nudge that interrupted them with self-driven
work prompts.
keep-alive: when lead is idle, nudge it to scan team status, review
PR queue, and proactively dispatch the next issue to free workers.
Test plan
pytest tests/test_nudge_coordinator.py(19 passed)Closes #223.