The Dagger container running generate_build_history.py may not always
reach the deployment server (network constraints on the Dagger engine).
Rather than aborting the entire publish-website pipeline, log the SSH
verbose output (already added in the previous debug commit) and return
an empty file list so Hugo still builds and rsync still deploys the
site — just without updated build-history pages.
This unblocks the cron deploy that has been failing since c259d2da.
Temporary: print verbose SSH output on failure to identify why the
connection fails from inside the dagger container.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Dagger mounts the secret file with 0600 but the parent directory may
get created with world-readable permissions, causing SSH to refuse
the key with exit 255.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
All other ssh/scp calls in the dagger module use explicit -i /root/.ssh/id_ed25519.
This one was missing it, causing exit 255 inside the dagger container.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Dagger parses .env directly and fails on multiline quoted values.
Move SSH_PRIVATE_KEY out of .env and export it from ~/.ssh/id_ed25519
in the wrapper instead.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Tracks consecutive failure count in .fail_count. On the 5th failure
for the same SHA, creates a Prio/High + State/Ready Codeberg issue.
Before creating, checks local .last_issue_sha and queries Codeberg
open issues to avoid duplicates.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
If the last deploy failed and origin/main has not advanced, opens a
Prio/High + State/Ready issue via tea with the failing SHA, commit link,
and captured deploy output. Skips duplicate issues (tracked by
.last_issue_sha). Cron interval changed to */5.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- agent_loop.py: create log dir with mode 0700 and enforce it on
existing dirs; open log files with mode 0600; chmod state file
to 0600 after every write. Prevents other local processes from
reading agent output (which may contain credential paths) or
tampering with the state file's pid field.
- ci/main.go (TestAndroidFirebase): replace
echo "$FIREBASE_SA_KEY" > /tmp/key.json
with bash process substitution
--key-file=<(echo "$FIREBASE_SA_KEY")
The key is now passed via a file descriptor — it never touches
disk, so it cannot be stranded by a failed gcloud auth call or
snapshotted into the Dagger layer cache.
- ci.yml / deploy.yml: add "Cleanup TLS credentials" step
(if: always()) at the end of every job that calls
setup_dagger_remote.sh. Removes /tmp/dagger-tls,
/tmp/stunnel-dagger.conf, /tmp/stunnel.pid from the self-hosted
runner after each job, so client certs do not accumulate between
job runs.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
flutter pub get is pure Dart — it never invokes Gradle. The mutable
gradle-cache volume mount caused the same execution-cache instability
we just fixed for the pub cache: Dagger sees a changed volume and
cache-misses pubGetLayer() on every run.
The Gradle cache stays in Base(), which is only used for steps that
actually build Android code.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The mutable flutter-pub-cache volume made the execution cache key unstable —
pub get cache-missed every run because the volume's mutable layer changed the
snapshot hash. Removing the volume lets Dagger snapshot packages inside the
execution-cache layer, which is stable and reclaimable via dagger prune.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
On some Android versions the path_provider Pigeon channel
('dev.flutter.pigeon.path_provider_android.PathProviderApi.getApplicationSupportPath')
is not ready when initDatabasePath() runs before runApp(). The existing code
already catches PlatformException there, leaving _dbPath null — but the
LazyDatabase callback called getApplicationSupportDirectory() a second time
without any protection, causing an unhandled crash on those devices.
Fix: extract _resolveDatabasePath() which retries three times with back-off
(100 ms → 300 ms → 600 ms) before re-throwing with a descriptive error
message. By the time the database is first accessed (after runApp()), the
channel is almost always available; if it still isn't, the CrashScreen is
shown with a clear explanation.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Only pick up issues created by guettli, guettlibot, or guettlibot2
to prevent the loop from acting on external/bot issues.
- Post an explanatory comment on the issue whenever the loop sets
State/Question (agent killed, no CI run, no push detected), so the
reason is visible without digging through cron logs. Closes#158.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Firebase CLI emits "A non-retryable error occurred." even for passing runs.
The grep -qwi 'error' triggered on this message despite gcloud exiting 0
and the result table showing Passed. The gcloud exit code, device-count,
and Passed checks are sufficient to detect real failures.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- agent_loop.py: agents now create an `issue-N-fix` branch and open a PR;
the loop discovers the PR via `fgj pr list`, tracks its CI run, squash-merges
on green, and falls back to the global-CI path if no PR exists (backward compat).
Adds `_find_pr_for_branch`, `_latest_ci_run_for_branch`, `_merge_pr` helpers.
- .forgejo/workflows/ci.yml: strip to the single fast `check` job only
(removes build-linux, deploy-playstore, publish-website).
- .forgejo/workflows/deploy.yml (new, replaces android-emulator-tests.yml):
scheduled hourly + workflow_dispatch; runs firebase tests, Play Store deploy,
Linux build/deploy, website publish; on completion sets CI/Full-Pass or
CI/Full-Fail label on the repo's DEPLOY_HEALTH_ISSUE tracking issue.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
If the agent exits immediately (e.g. rate-limit), the loop was closing the
pending issue against the *previous* CI run, which was still green.
Fix: record the latest CI run ID when an issue agent starts. If the run ID
hasn't changed when the agent exits, the agent pushed nothing → set
State/Question instead of closing.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
flutter_markdown 0.7.7+1 has been discontinued in favour of
flutter_markdown_plus. Switch the dependency and update both import
sites.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Capture gcloud auth stderr separately and fail on unexpected output;
ignore the two known informational lines ("Activated service account
credentials for: [...]" and "Updated property [core/project].") while
keeping a strict "fail if unknown stderr" check for anything else.
- Replace the narrow pattern grep (non-retryable error|infrastructure_failure|
test execution failed) with a broad whole-word case-insensitive grep for
'error', so any infrastructure or Firebase error in the output causes CI
failure.
- Verify that the number of device result rows in the result table matches
the expected device count (1), so a silent test-run failure cannot slip
through.
- Add scripts/test_firebase_check.sh with 18 unit tests for the three new
bash patterns (auth stderr filter, error-word detection, device count).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
All production secrets (SSH key, Android keystore, Play Store config,
Firebase service account) are already typed as dagger.Secret and injected
via WithMountedSecret / WithSecretVariable. Add a Secrets section to
DAGGER.md to make this explicit.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The gradle-cache volume was mounted without an owner, so the root-owned
volume caused "Permission denied" when the ci user tried to create
gradle-8.14-all.zip.lck during bundleRelease. Add Owner: "ci" to all
three WithMountedCache calls so the ci user can write to the caches.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add -qq to apt-get update/install in Dagger toolchain to suppress
verbose package-list output (hundreds of lines on cold cache)
- Wrap sdkmanager in silent-on-success pattern — only shows output
on failure, like the build_runner and flutter pub get steps
- Set debug = warning in stunnel config to suppress LOG5 (info/notice)
startup lines while keeping LOG4 (warning) and above
- Add org.gradle.welcome=never to android/gradle.properties to
suppress the "Welcome to Gradle N.NN!" banner
- Filter SKIPPED Gradle tasks, Gradle Daemon startup messages, and
gcloud support-page promo lines in run_firebase_test.sh
Errors and warnings are preserved in all cases.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The cirruslabs/flutter:3.41.6 image already has UID 1000 assigned to
another user, so `useradd -u 1000` exits with code 4 ("UID not unique")
and the ci user is never created. Dagger then fails to resolve `owner:
"ci"` on subsequent WithDirectory calls. Removing the explicit UID lets
useradd pick the next available one.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Create a non-root user 'ci' (UID 1000) in the Dagger toolchain container,
transfer ownership of the Flutter SDK and Android SDK to that user, and
switch to it with WithUser("ci"). Update all cache mount paths from /root/
to /home/ci/ and set Owner: "ci" on every WithDirectory call so Flutter
can write build output. Flutter emits a strong warning when run as root;
this change eliminates that warning by running the tool as a regular user.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
On some Android devices (e.g. Android S1RXS32.50-13-25) the WorkManager
platform channel fails to connect at startup, throwing
PlatformException(channel-error, ...). registerBackgroundSync() now catches
PlatformException and MissingPluginException (plus any other unexpected
failure) and silently disables background sync rather than crashing the app.
Test added: test/unit/background_sync_test.dart verifies the function
completes without throwing in the unit-test environment (where the native
plugin is absent).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
`tea api` exits 0 even on 401 responses, so `_close_issue` and
`_set_labels` appeared to succeed but did nothing. Issues were never
actually closed, causing them to be picked up again every cron tick.
Switch all write operations (close issue, set labels) and issue-list
reads to `fgj`, which has proper authentication. Keep `tea api` only
for CI run fetches where `fgj` times out (504). Add ~/go/bin to the
cron PATH so fgj is found.
Also add an error check in `_tea_get` for API-level error responses,
and strip State/InProgress when closing an issue.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Two fixes:
1. notification_service.dart: initNotifications() now catches
MissingPluginException (and any other init failure) so the app no
longer crashes when flutter_local_notifications is unavailable on
some Android devices. _initialized tracks success; showNewMailNotification
skips the plugin call when it never initialised.
2. crash_screen.dart: "Report Issue on Codeberg" no longer puts the full
report in the URL query string. Long stack traces exceeded browser
URL-length limits and caused "create issue failed". The URL now
carries only the pre-filled title; the user copies the full report
via "Copy to Clipboard" and pastes it in the issue body.
Tests added:
- test/unit/notification_service_test.dart: verifies initNotifications()
completes without throwing when the plugin channel is unavailable.
- test/widget/crash_screen_test.dart: verifies the Codeberg URL contains
the title but no &body= parameter.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Previously build_runner compiled separately for each setup() variant
(checkSrc, backendSrc, integrationSrc, etc.) since their differing
source inputs produced distinct Dagger cache keys. CheckMocks also ran
build_runner twice: once inside setup() and again explicitly — and the
second run always compared two freshly-generated outputs, so stale mocks
in the repo were never detected.
Introduce codegenBase() that runs build_runner on the minimal common
source (lib/, test/, assets/, pubspec.*) excluding committed generated
files. All setup() calls now share this single Dagger cache entry, so
build_runner compiles only once per pipeline run instead of once per
source variant.
Fix CheckMocks to start from pubGetLayer() + committed source (including
any stale *.mocks.dart), commit that state as the git baseline, then run
build_runner once. The subsequent git diff now correctly detects stale
mocks in the repository, matching the behaviour of check_mocks_fresh.sh.
Also update Graph() to reflect the new codegenBase node.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Previously issue agents were instructed to close the issue via prompt text
immediately after pushing. If CI then failed, the issue was already closed.
Now the loop tracks a pending_issue across cron ticks:
- When an agent finishes (issue or ci-fix), the issue number is extracted
from state before it is cleared.
- If CI is still running, a "pending-ci" state preserves the issue number.
- If CI fails, the ci-fix agent is started with the issue number in state
so it survives the fix cycle.
- Once CI passes, _close_issue() is called from Python — never by the agent.
The agent prompt no longer instructs the agent to close the issue.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add `---------------------- Starting YYYY-MM-DD HH:MMZ` header at each run
- Remove `[agent_loop]` prefix from all output lines
- Show full Codeberg URL for CI runs instead of bare run ID
- Show full issue URL and title when referencing issues
- Store issue_title in state file so "still running" messages include the title
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
State/Ready → State/InProgress is already set by agent_loop.py before
the agent starts. Update AGENTS.md to reflect that agents invoked via
the loop must not set InProgress themselves (only manual workflows need
to). Also fix TestMain tests that called main() directly, which caused
argparse to consume sys.argv; they now call _run_loop() instead.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Previously, `gcloud firebase test android run` could exit 0 while printing
"A non-retryable error occurred." in its output. The old check
`&& echo "$out" || { exit 1; }` only caught non-zero exit codes, and the
success grep `'Passed|passed|test cases'` was too broad — "test cases" can
appear in Firebase output before the error, giving a false positive.
The fix captures gcloud's exit code explicitly via `rc=$?`, adds an explicit
error-string check for known Firebase failure phrases (non-retryable error,
infrastructure_failure, test execution failed), and tightens the success
pattern to `'Passed|passed'` only.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Two bugs caused the crash-at-startup report:
1. CrashScreen used the widget's build context (above its own MaterialApp)
for ScaffoldMessenger.of() in button callbacks. When the screen is the
root widget — the runApp() path after a startup crash — there is no
ScaffoldMessenger above it, so both 'Copy to Clipboard' and 'Report Issue
on Codeberg' crashed with a null check error. Fix: wrap Scaffold.body in
Builder to obtain a context that is a descendant of the Scaffold.
2. path_provider_android 2.2.21 updated to Pigeon 26, which causes a
channel-error on startup for some Android devices. Pin to <2.2.21
(resolves to 2.2.20, which uses the stable pre-Pigeon-26 implementation).
Additionally, make initDatabasePath() catch PlatformException so a
channel error at the very start of main() no longer hard-crashes the app;
_openConnection()'s lazy fallback retries after runApp() completes.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace curl-based install of dagger/task with a hard check that
fails immediately if any tool is missing from the runner image,
pointing to .forgejo/Dockerfile as the fix location.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Based on ghcr.io/catthehacker/ubuntu:go-24.04 with stunnel4,
netcat-openbsd, dagger v0.20.8 and task v3.48.0 baked in so
nothing is downloaded during CI runs.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
gcloud exits 0 even when no tests ran. Add a post-check that greps
the output for 'Passed/passed/test cases' and fails explicitly if
none are found, so 'no test case results' turns the CI red.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Previously setUpAll() fell back to 127.0.0.1 defaults when env vars
were absent, causing Firebase Test Lab to report '0 test case results'
instead of a clear failure. Now it calls fail() immediately with the
list of missing variables.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace apt-get install with a hard check — if the packages are missing
the job fails immediately with a clear error. Avoids flaky failures when
archive.ubuntu.com is unreachable.
Install once on the runner: sudo apt-get install -y stunnel4 netcat-openbsd
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
pubspec.lock was incorrectly gitignored — this is a Flutter app, not a
package, so the lockfile should be committed for reproducible builds.
Without it, CI resolved drift to its minimum (2.20.3) which constrains
sqlite3 to 2.x, causing dart analyze to disagree on whether
Database.close() exists vs the local environment using 3.3.1.
Also pins sqlite3: ^3.1.5 explicitly in pubspec.yaml as belt-and-
suspenders so the constraint is visible without reading the lockfile.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The default Firebase Test Lab bucket is in a Google-managed project so
project-level IAM grants have no effect on it. Use sharedinbox-ftl-results
which is in sharedinbox-496103 where the service account has storage.admin.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
WithEnvVariable(CACHE_BUSTER, time.Now()) ensures gcloud firebase test
always runs fresh rather than returning a cached result from a prior run.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
