OpenCode's own snapshot tracker should not `git add` an ArchiveBox
collection. On a real `/data` project it traversed archived nested
repositories, failed on their Git state, and left the agent request busy
after the model had replied.
Generate fresh ArchiveBox-owned OpenCode config with `snapshot: false`.
Existing OpenCode config remains entirely user-owned and untouched.
Verification:
- real config/server tests pass
- DigestBox `/data` smoke selected `opencode/big-pickle`, replied
correctly, exited cleanly in 79s, and logged no snapshot
traversal/failure.
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
OpenCode's snapshot tracker was `git add`-ing the ArchiveBox collection,
failing on nested archived repos and leaving agent requests busy after
the model replied. Disables filesystem snapshots for generated OpenCode
configs by setting `snapshot: false`, and updates tests to assert the
new config. Existing user-owned configs are left untouched.
<sup>Written for commit e7ff7cdad2.
Summary will update on new commits.</sup>
<a
href="https://cubic.dev/pr/ArchiveBox/ArchiveBox/pull/1868?utm_source=github"
target="_blank" rel="noopener noreferrer"
data-no-image-dialog="true"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://www.cubic.dev/buttons/review-in-cubic-dark.svg"><source
media="(prefers-color-scheme: light)"
srcset="https://www.cubic.dev/buttons/review-in-cubic-light.svg"><img
alt="Review in cubic"
src="https://www.cubic.dev/buttons/review-in-cubic-dark.svg"></picture></a>
<!-- End of auto-generated description by cubic. -->
Follow-up to #1866 addressing both Cubic findings before release:
- recognize schema-only OpenCode JSONC with comments and trailing commas
- exercise the intended preservation branch with valid JSON
- also verify administrator-selected JSONC remains byte-for-byte
unchanged
Verification: focused config tests pass (5 passed).
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Fixes detection of schema-only OpenCode JSONC configs so comments,
trailing commas, and escaped keys no longer prevent the default model
from being added.
- Replaces strict JSON parsing with a JSONC-aware parser that strips
comments and trailing commas before checking whether a config only sets
`$schema`.
- Writes the default config as JSONC text instead of JSON.
- Expands tests to cover plain JSON, JSONC with comments and trailing
commas, and escaped schema keys.
<sup>Written for commit c1358a2726298e20db1a8697647338baa21e555f.
Summary will update on new commits.</sup>
<a
href="https://cubic.dev/pr/ArchiveBox/ArchiveBox/pull/1867?utm_source=github"
target="_blank" rel="noopener noreferrer"
data-no-image-dialog="true"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://www.cubic.dev/buttons/review-in-cubic-dark.svg"><source
media="(prefers-color-scheme: light)"
srcset="https://www.cubic.dev/buttons/review-in-cubic-light.svg"><img
alt="Review in cubic"
src="https://www.cubic.dev/buttons/review-in-cubic-dark.svg"></picture></a>
<!-- End of auto-generated description by cubic. -->
Fixes the DigestBox agent path end to end:
- use the ArchiveBox collection root as the default OpenCode workdir
- configure OpenCode's free `opencode/big-pickle` model for
fresh/schema-only state while preserving administrator config
- remove the redundant two-second version gate and three-second browser
health poll that falsely hid a working server and generated continuous
traffic
- retire the private health endpoint and its tests
Verification: `uv run pytest archivebox/tests/test_opencode_agent.py -q`
(20 passed)
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Fixes OpenCode agent startup so fresh installs get working defaults and
a healthy server no longer appears unavailable.
- Uses the ArchiveBox collection root as the default OpenCode workdir
instead of an unused subdirectory.
- Writes the `opencode/big-pickle` default model only when
`opencode.jsonc` is missing or contains only `$schema`; existing admin
config is left untouched.
- Removes the two-second version gate and three-second browser health
poll that falsely marked healthy servers unavailable and generated
continuous traffic.
- Deletes the private `/_archivebox/health` endpoint and its tests.
<sup>Written for commit 9a3f058f73.
Summary will update on new commits.</sup>
<a
href="https://cubic.dev/pr/ArchiveBox/ArchiveBox/pull/1866?utm_source=github"
target="_blank" rel="noopener noreferrer"
data-no-image-dialog="true"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://www.cubic.dev/buttons/review-in-cubic-dark.svg"><source
media="(prefers-color-scheme: light)"
srcset="https://www.cubic.dev/buttons/review-in-cubic-light.svg"><img
alt="Review in cubic"
src="https://www.cubic.dev/buttons/review-in-cubic-dark.svg"></picture></a>
<!-- End of auto-generated description by cubic. -->
Removes the issue-opened duplicate detection workflow entirely. It is
not part of product CI and currently depends on separately funded
Anthropic API credits. No replacement, fallback, or failure-suppression
path is added.\n\nValidation: `git diff --check`; `uv run prek run
--all-files`.
The rc419 DigestBox test showed that health-based replacement still
killed OpenCode during legitimate initial UI load: health became
temporarily unavailable for more than five seconds while the child was
busy, causing repeated process replacement and 502s.
This removes the health-latency heuristic entirely. ArchiveBox now
distinguishes ready from merely running, trusts a ready live owned PID,
serializes cold starts until readiness, and only replaces children that
exited or never became ready. Stale tabs still recover after an
ArchiveBox/container restart because process state resets and the
detached health wake starts a new child.
This is also a net cleanup: 38 lines removed from the prior lifecycle
implementation.
Verification:
- `uv run pytest archivebox/tests/test_opencode_agent.py -q` (19 passed)
- focused real startup/readiness/proxy/shutdown tests (5 passed)
- `uv run prek run --all-files`
- independent gpt-5.6-medium lifecycle review clean
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Removes the health-latency heuristic that replaced OpenCode processes
during legitimate initial UI loads, causing 502s. Ready owned processes
are now trusted without probing, and startup health checks are bounded
so they can't run past the startup deadline.
- Only replaces children that exited or never became ready.
- Cold starts are serialized until readiness; stale tabs still recover
after a restart.
<sup>Written for commit 57e7820a0b.
Summary will update on new commits.</sup>
<a
href="https://cubic.dev/pr/ArchiveBox/ArchiveBox/pull/1862?utm_source=github"
target="_blank" rel="noopener noreferrer"
data-no-image-dialog="true"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://www.cubic.dev/buttons/review-in-cubic-dark.svg"><source
media="(prefers-color-scheme: light)"
srcset="https://www.cubic.dev/buttons/review-in-cubic-light.svg"><img
alt="Review in cubic"
src="https://www.cubic.dev/buttons/review-in-cubic-dark.svg"></picture></a>
<!-- End of auto-generated description by cubic. -->
The rc418 live DigestBox restart test exposed OpenCode process flapping
under iframe load: a single 2-second health timeout caused
`_ensure_opencode` to kill an otherwise live owned process, after which
the reloading iframe repeated the cycle.
This gives an owned process a bounded 5-second health recovery window
before replacing it. A real SIGSTOP/SIGCONT test proves that a
transiently unavailable process keeps the same PID, while the existing
real unhealthy-process test still proves eventual replacement.
Verification:
- `uv run pytest archivebox/tests/test_opencode_agent.py -q` (18 passed)
- `uv run prek run --all-files`
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Stops OpenCode from restarting on a single 2-second health timeout,
which caused process flapping under iframe load. An owned process now
gets a bounded 5-second health recovery window before being replaced, so
a transiently unavailable process keeps the same PID. Proxy and
event-stream requests no longer block on the recovery check when the
process is already running.
- Adds a SIGSTOP/SIGCONT test proving a transiently stopped process
keeps its PID, while the existing unhealthy-process test still verifies
eventual replacement.
- Adds tests proving proxy requests skip the recovery check while the
process is running and wait for readiness when it is not.
<sup>Written for commit 650697f3c2.
Summary will update on new commits.</sup>
<a
href="https://cubic.dev/pr/ArchiveBox/ArchiveBox/pull/1861?utm_source=github"
target="_blank" rel="noopener noreferrer"
data-no-image-dialog="true"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://www.cubic.dev/buttons/review-in-cubic-dark.svg"><source
media="(prefers-color-scheme: light)"
srcset="https://www.cubic.dev/buttons/review-in-cubic-light.svg"><img
alt="Review in cubic"
src="https://www.cubic.dev/buttons/review-in-cubic-dark.svg"></picture></a>
<!-- End of auto-generated description by cubic. -->
## Summary
- monitor the proxied OpenCode health endpoint from the stable
ArchiveBox parent page
- reload the iframe once after an unavailable -> healthy transition
- trigger an immediate check again when the browser comes online or the
tab becomes visible
## Why
A real RC416 -> RC417 DigestBox restart left the already-open OpenCode
1.17.14 SPA showing its fatal error page against the newly started
OpenCode 1.17.15 server. A full page reload recovered immediately. The
parent ArchiveBox document survives the container restart, so it can
perform that one iframe reload automatically after the server is healthy
again.
## Verification
- `test_opencode_agent_superuser_gets_admin_wrapper` passes with
assertions for the recovery monitor
- focused pre-commit hooks pass
- real stale-tab restart reproduction will be repeated on DigestBox
after release
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Reloads stale OpenCode iframes after the ArchiveBox server recovers from
a restart, so the previously fatal error page is replaced automatically.
The parent page polls the OpenCode health endpoint every 3 seconds,
reloading the iframe once when the server transitions from unavailable
to healthy or when the server version changes, and also re-checks
immediately when the browser comes online or the tab becomes visible.
- Health monitoring uses a separate endpoint that reports status without
starting the server; recovery restarts a running but unhealthy OpenCode
process, and polling skips while hidden when the server is down.
**Verification**
- `test_opencode_agent_superuser_gets_admin_wrapper` passes with
assertions for the recovery monitor.
- Focused pre-commit hooks pass.
<sup>Written for commit a33b333b6a.
Summary will update on new commits.</sup>
<a
href="https://cubic.dev/pr/ArchiveBox/ArchiveBox/pull/1860?utm_source=github"
target="_blank" rel="noopener noreferrer"
data-no-image-dialog="true"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://www.cubic.dev/buttons/review-in-cubic-dark.svg"><source
media="(prefers-color-scheme: light)"
srcset="https://www.cubic.dev/buttons/review-in-cubic-light.svg"><img
alt="Review in cubic"
src="https://www.cubic.dev/buttons/review-in-cubic-dark.svg"></picture></a>
<!-- End of auto-generated description by cubic. -->
## Summary
- return OpenCode SSE headers before waiting for a stopped server to
restart
- run reconnect-driven OpenCode startup in a non-thread-sensitive worker
inside the async stream
- remove the obsolete global `ThreadSensitiveContext` monkeypatch now
that asgiref 3.12.1 implements non-blocking shutdown upstream
- add a real ASGI regression test that holds the real startup lock and
verifies headers arrive before restart can finish
## Why
After deploying RC416, a stale `/admin/agent` tab successfully recovered
across the container restart, but Daphne logged three
`CurrentThreadExecutor already quit or is broken` exceptions while early
EventSource reconnects were cancelled during synchronous OpenCode
startup. Normal proxy traffic subsequently returned 200 and no data was
lost, but the request cleanup path was not clean.
## Verification
- `uv run pytest archivebox/tests/test_opencode_agent.py -q` — 15 passed
- focused SSE tests after final refactor — 2 passed
- `uv run prek run --all-files` — passed
- no OpenCode server remained after fixture teardown
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Makes OpenCode SSE reconnects return headers before the stopped server
finishes restarting, so early EventSource reconnects no longer cancel
while startup runs on a blocked ASGI request thread. This removes the
`CurrentThreadExecutor already quit or is broken` exceptions Daphne
logged after container restarts.
- Runs reconnect-driven OpenCode startup in a non-thread-sensitive
worker inside the async stream instead of blocking the request thread.
- Surfaces OpenCode startup failures as an error event on the SSE stream
instead of returning an error response.
- Removes the obsolete global `ThreadSensitiveContext` monkeypatch from
`archivebox/core/asgi.py`; asgiref 3.12.1 handles non-blocking shutdown
upstream.
- Adds an ASGI regression test that holds the real startup lock,
verifies headers arrive before restart can finish, avoids starting a
background server, and joins the startup worker before teardown.
<sup>Written for commit 7d7e219962.
Summary will update on new commits.</sup>
<a
href="https://cubic.dev/pr/ArchiveBox/ArchiveBox/pull/1859?utm_source=github"
target="_blank" rel="noopener noreferrer"
data-no-image-dialog="true"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://www.cubic.dev/buttons/review-in-cubic-dark.svg"><source
media="(prefers-color-scheme: light)"
srcset="https://www.cubic.dev/buttons/review-in-cubic-light.svg"><img
alt="Review in cubic"
src="https://www.cubic.dev/buttons/review-in-cubic-dark.svg"></picture></a>
<!-- End of auto-generated description by cubic. -->
