Gitea 28 still defaults the setting to false, so dropping the explicit
pin changes no behaviour and removes its startup warning. Existing
installs keep whatever they already have; the startup script never
touched this key.
- rootfs/tmp/etc/gitea/app.ini: remove DISABLE_QUERY_AUTH_TOKEN and its explanatory comment from [security]
- TODO.AI.md: note the removal in the Gitea 28 migration entry
Claude-Session: https://claude.ai/code/session_01Kae22tmhzhmfHyqc34PZbx
Audited the template against Gitea 28.0.0 (source plus the binary's
own startup log) and fixed everything it flags. Existing configs are
migrated on startup without touching user values.
- rootfs/tmp/etc/gitea/app.ini: ALLOWED_HOST_LIST "*" -> private,loopback with explicit EGRESS_MODE=lax; [migration] -> [migrations] with ALLOWED_HOST_LIST; drop deprecated/dead keys and the non-existent [notify.email]; add WORK_PATH and LOCAL_ROOT_URL; DISABLE_REGISTRATION=true
- rootfs/usr/local/etc/docker/init.d/08-gitea.sh: add __gitea_ini_get/__gitea_ini_unset and migrate existing configs (invalid "*" host list, [webhook] key moved to [security], [migrations] host list, EGRESS_MODE)
- TODO.AI.md: record the audit, the verification, and the behaviour changes needing owner confirmation
Claude-Session: https://claude.ai/code/session_01Kae22tmhzhmfHyqc34PZbx
Gitea now always listens on plain http while ROOT_URL carries the
public https origin, per Gitea's reverse-proxy docs. Fixes fresh
containers failing to start (PROTOCOL = https with no cert) and
passkey origin mismatches behind a proxy.
- rootfs/usr/local/etc/docker/init.d/08-gitea.sh: add __gitea_ini_set helper; stamp [server] PROTOCOL = http, ROOT_URL from new GITEA_ROOT_URL (trailing slash), and [session] COOKIE_SECURE from ROOT_URL scheme on every startup
- rootfs/tmp/etc/gitea/app.ini: default PROTOCOL to http instead of the public protocol placeholder
- README.md: document GITEA_PROTO as the public protocol (default https), GITEA_ROOT_URL, and reverse-proxy/passkey requirements; compose example uses https
- TODO.AI.md: record the fix and log the generated __create_ssl_cert crash
Claude-Session: https://claude.ai/code/session_01Kae22tmhzhmfHyqc34PZbx
- Replace hardcoded `PROTOCOL = http` with `REPLACE_SERVER_PROTO` in the `[server]` section
- Keeps PROTOCOL consistent with ROOT_URL, which already uses the placeholder
- Lets the entrypoint set the protocol at runtime instead of forcing http
AGENTS.md
rootfs/tmp/etc/gitea/app.ini
Gitea config: `ALLOWED_HOST_LIST` was misplaced under `[webhook]` — the current
config-cheat-sheet places it under `[security]` (default `external`), where it
gates outbound webhook/OAuth2 calls. Left at its `[webhook]` default of unset
(no such key there), `[security] ALLOWED_HOST_LIST` silently fell back to
`external`, blocking internal-facing calls triggered by API actions like org
creation and surfacing as an opaque 502 through the reverse proxy. Also pinned
`DISABLE_QUERY_AUTH_TOKEN=false` explicitly, since Gitea flips its default to
`true` in 1.23 (deprecated in 1.24) and this image always builds against the
latest Gitea release — leaving it unset would silently downgrade `?token=`
API calls to anonymous on the next image rebuild.
Runtime/entrypoint: hardened `grep`/`type -t | grep` calls with `--` across
the entrypoint function library and init.d scripts to stop values starting
with `-` from being parsed as flags; guarded the entrypoint and `__no_exit`
monitor-loop PID-reuse checks with a cmdline marker (PID namespaces reset on
`docker restart` but `/run` persists, so a recorded PID can coincidentally be
reused by an unrelated process and falsely appear "still running"); added a
stale `/tmp/docker.pid` cleanup before each dockerd start attempt for the
same reason; added `NO_COLOR`-aware plain-text fallbacks for emoji status
banners; renamed the `su_cmd` helper to `__su_cmd` for naming consistency
with other private functions; added `fuse-overlayfs` as the Docker-in-Docker
storage driver.
act_runner: removed the legacy single "gitea"-named runner registration and
daemon start in `zz-act_runner.sh` — `start-runners` already owns all runner
registration/count via `RUNNERS_START`, and running both duplicated runners.
Registration now targets `127.0.0.1` instead of the detected external IPv4
address, which is transient/wrong under Docker-in-Docker networking and
caused "no route to host" registration failures. Long-running background
jobs (`cache-server`, `start-runners`) now redirect stdout/stderr to real log
files and are `disown`ed instead of inheriting the `__post_execute` pipe —
otherwise the `tee` reading that pipe never sees EOF and `__run_start_script`
hangs forever waiting on a process that never exits. `start-runners` gained a
version-stamp header and builds `RUNNER_LABELS` from an array instead of one
long string for readability; its `ERR` trap and ports list now respect
`NO_COLOR`.
- rootfs/tmp/etc/gitea/app.ini: move `ALLOWED_HOST_LIST` to `[security]`; add explicit `DISABLE_QUERY_AUTH_TOKEN=false`
- rootfs/tmp/etc/docker/daemon.json: add `storage-driver: fuse-overlayfs`
- rootfs/usr/local/bin/entrypoint.sh: version bump; `grep --` hardening; cmdline-marker PID-reuse guard; `exit 0` instead of bare `exit`
- rootfs/usr/local/bin/start-runners: add version-stamp header; `NO_COLOR`-aware ERR trap; build `RUNNER_LABELS` from an array
- rootfs/usr/local/etc/docker/functions/entrypoint.sh: version bump; `grep --` hardening across helpers; `__no_exit` monitor-loop cmdline-marker guard; `NO_COLOR`-aware service banners; rename `su_cmd` to `__su_cmd`
- rootfs/usr/local/etc/docker/init.d/05-dockerd.sh: version bump; `NO_COLOR`-aware messages; stale `/tmp/docker.pid` cleanup before start; `symlink`/`su_cmd` calls updated to `__symlink`/`__su_cmd`; `grep --` hardening; add `storage-driver` to both daemon.json heredocs
- rootfs/usr/local/etc/docker/init.d/08-gitea.sh: version bump; `NO_COLOR`-aware messages (including stale-PID-file cleanup); `grep --` hardening; `su_cmd` call updated to `__su_cmd`
- rootfs/usr/local/etc/docker/init.d/zz-act_runner.sh: version bump; remove legacy duplicate runner registration/daemon start; register against `127.0.0.1`; redirect and disown long-running background jobs to prevent pipe hangs; `NO_COLOR`-aware messages
Turned on the act_runner Actions cache instead of leaving it disabled.
The standalone `cache-server` process was already being launched
unconditionally by zz-act_runner.sh but had no `external_secret`, so it
was refusing to start (the binary requires a non-empty secret even
though its own `cache.enabled` field is unused). Wired a shared,
randomly generated secret through the existing REPLACE_* templating
pattern so the "gitea" runner, the cache-server, and the extra
runner-N daemons spawned by start-runners all share one cache backend.
- rootfs/tmp/etc/act_runner/default_config.yaml: cache.enabled: true;
added external_server (pointed at the local cache-server) and
external_secret placeholders
- rootfs/tmp/etc/act_runner/cache_server.yaml: cache.enabled: true;
added external_secret placeholder (required for the binary to start)
- rootfs/usr/local/etc/docker/init.d/zz-act_runner.sh: added
__gen_cache_secret() (mirrors __gen_auth_token's persisted-token
pattern, stored at $CONF_DIR/tokens/cache_secret); substitutes
REPLACE_RUNNER_CACHE_SECRET into both the runner config and the
cache-server config; exports RUNNER_CACHE_HOST/PORT/SECRET for
start-runners; bumped version stamp
- rootfs/usr/local/bin/start-runners: generates a shared
runners-cache.yaml from the exported cache env vars and passes
--config to both `act_runner register` and `act_runner daemon` for
every runner-N instance, so they use the same external cache server
instead of an unshared per-process local cache
- TODO.AI.md: logged pre-existing script-lint findings (missing `--`
before grep queries throughout zz-act_runner.sh, a UUOC in
start-runners, an unpinned/stale docker.yaml CI workflow, and
forbidden OCI labels in the Dockerfile) surfaced incidentally by the
lint pass for this change but out of scope for it
Forks should get the same full set of units as new repos since the
intent is a full GitHub migration. Matches DEFAULT_REPO_UNITS exactly.
- rootfs/tmp/etc/gitea/app.ini: add DEFAULT_FORK_REPO_UNITS with all units
rootfs/tmp/etc/gitea/app.ini
Add DEFAULT_REPO_UNITS to [repository] in app.ini with repo.actions
included so every new repo has Actions enabled out of the box.
Users can still disable it per-repo — opt-out rather than opt-in.
- rootfs/tmp/etc/gitea/app.ini: add DEFAULT_REPO_UNITS with repo.actions included
rootfs/tmp/etc/gitea/app.ini
default_config.yaml had cache.enabled=true but no external_secret,
dir, host, or port configured. act_runner refuses to start the cache
server without external_secret, logging an error on every boot.
cache_server.yaml already has cache disabled — align default_config
to match.
- rootfs/tmp/etc/act_runner/default_config.yaml: cache.enabled false
rootfs/tmp/etc/act_runner/default_config.yaml
SSL interception on this network presents certificates without SANs
for the target hostname, breaking git clone operations during Gitea
migrations. Add migration and git-level TLS bypass settings.
- rootfs/tmp/etc/gitea/app.ini: add [migration] section with SKIP_TLS_VERIFY=true
- rootfs/tmp/etc/gitea/app.ini: add [git.config] http.sslVerify=false
rootfs/tmp/etc/gitea/app.ini
Resolve all deprecation warnings emitted at runtime by updating
app.ini to use current key names and locations.
- rootfs/tmp/etc/gitea/app.ini: move X_FRAME_OPTIONS from [cors] to [security] (deprecatedSetting v1.26.0)
- rootfs/tmp/etc/gitea/app.ini: remove [picture].DISABLE_GRAVATAR (deprecatedSettingDB since v1.18, use admin panel)
- rootfs/tmp/etc/gitea/app.ini: remove [picture].ENABLE_FEDERATED_AVATAR (deprecatedSettingDB since v1.18, use admin panel)
- rootfs/tmp/etc/gitea/app.ini: rename [lfs].LFS_CONTENT_PATH → PATH (was silently ignored; correct key is PATH)
- rootfs/tmp/etc/gitea/app.ini: rename [git].MAX_GIT_DIFF_LINE_CHARACTER_COUNT → MAX_GIT_DIFF_LINE_CHARACTERS (correct struct field name)
- rootfs/tmp/etc/gitea/app.ini: remove LOG_ROTATE/DAILY_ROTATE/MAX_DAYS from [log] root section (only valid in file-writer sub-sections; dead config with MODE=console)
.claude/settings.local.json
rootfs/tmp/etc/gitea/app.ini
Three more bugs found during full component verification:
1. RUNNER_CONFIG_DEFAULT pointed to config.yaml but the actual
template filename is default_config.yaml — gitea named runner
never registered because the file-existence check always failed.
2. CACHE_CONFIG_FILE pointed to cache.yaml but the actual template
filename is cache_server.yaml — cache-server launch always
skipped silently.
3. act_runner v1.0.6 requires cache.external_secret on both the
cache-server and any runner using external_server; neither config
had it, so cache-server exited immediately and runner registration
failed when external_server was present in default_config.yaml.
4. Registration log redirect was 2>/dev/stdout >>"$log" (stderr went
to original stdout, not the log); corrected to >>"$log" 2>&1.
- rootfs/usr/local/etc/docker/init.d/zz-act_runner.sh:
- RUNNER_CONFIG_DEFAULT: config.yaml → default_config.yaml
- CACHE_CONFIG_FILE: cache.yaml → cache_server.yaml
- fix registration log redirect: >>"$RUNNER_LOG_FILE" 2>&1
- rootfs/tmp/etc/act_runner/default_config.yaml: remove
external_server (no shared secret configured; each runner uses
its own internal cache)
- rootfs/tmp/etc/act_runner/cache_server.yaml: set enabled: false
(cache-server requires external_secret; disabled until a proper
shared-secret setup is added)
Verified: 6 daemons on fresh start (1 gitea named + 5 runners), all
with 22 labels; clean reconnect on restart with no re-registration.
rootfs/tmp/etc/act_runner/cache_server.yaml
rootfs/tmp/etc/act_runner/default_config.yaml
rootfs/usr/local/etc/docker/init.d/zz-act_runner.sh
Runner registration was broken by three compounding bugs:
1. Token generated at script init time (before gitea was ready)
2. gitea CLI missing --work-path/--custom-path flags, writing fatal
log messages to stdout which got captured as the token value
3. Token assignment inside a piped subshell didn't propagate back
to the parent shell, leaving SYS_AUTH_TOKEN empty at runtime
4. INSTALL_LOCK=false in app.ini template caused gitea to start in
install-wizard mode, making generate-runner-token always fail
- rootfs/tmp/etc/gitea/app.ini: set INSTALL_LOCK=true so gitea
auto-initializes the SQLite DB on first run
- rootfs/usr/local/etc/docker/init.d/zz-act_runner.sh:
- defer SYS_AUTH_TOKEN: initialize to empty at global scope,
generate in __run_pre_execute_checks after gitea is confirmed up
- add --work-path/--custom-path to gitea CLI call; filter output
with grep -oE '[A-Za-z0-9]{20,}' to extract only the token
- guard token generation on INSTALL_LOCK=true in app.ini
- read token back from $CONF_DIR/tokens/system after the piped
__run_pre_execute_checks call returns (subshell escape)
Tested: fresh start registers all 5 runners with full 22-label set;
restart skips re-registration and reconnects all 5 daemons cleanly.
rootfs/tmp/etc/gitea/app.ini
rootfs/usr/local/etc/docker/init.d/zz-act_runner.sh
casjay
