Repository navigation
Reconcile feature/free-tier-fleet with main (+ instruction.test.ts $TMPDIR self-pollution fix) - #1520
Merged
jeonghun-jj-lee merged 9 commits intoSep 24, 2026
Conversation
* fix(packaging): enforce known fixes before CI advisory * fix(app): #1466 restore parent Files Changed invalidation * test(app): cover all #1466 file edit events * fix(app): #1467 bound mirror and session warming * fix(packaging): migrate #832 shared-state guard * fix(app): guard session warm scheduler keys * fix(app): repair #1466 review findings --------- Co-authored-by: amicode-ci <ci@amicode.local>
* ci: publish release VSIXes to Open VSX Registry * ci: decouple Open VSX publish from Marketplace step outcome
…h every real boot (current-main slice, manifest-clean) (#1473) * fix(app): #1458 — the empty-workspace landing's titlebar portals rendered SessionChatsDropdown/StatusPopoverV2 (per-directory sync-context readers) at the "/" route, which never sits inside a SyncProvider; at real-boot timing the no-directory window renders the landing briefly on EVERY boot and the useSync() throw killed the whole route tree — including the reactive landing effect, so no draft was ever created and the app stayed frozen at "/". Portals removed, content kept; a guard test pins the landing module free of directory-scoped context components (the unit-suite trap: mocked contexts pass while the real boot crashes — caught live by the e2e rig). Manifest: files + classification updated (drift gate PASS). * fix(app): #1458 follow-up — the guard test matched raw text and tripped on the fix's own explanatory comment (which names the removed components); assert the import graph instead. All green: 2 pass, drift gate PASS. * fix(app): render empty landing without directory context --------- Co-authored-by: amicode-ci <ci@amicode.local>
…59b, current-main slice) (#1474) * fix(app): #1459b — retrying dynamic imports, the tunnel-truncation class (current-main slice). The fleet's ssh tunnel restarts on network transitions (launchd KeepAlive) and kills every in-flight tunneled response MID-STREAM; a lazily-imported chunk crossing a restart gets a cleanly-truncated body (observed live: 2,884 of 28,497 bytes) and Chromium's module map caches the failure FOR THE DOCUMENT'S LIFE — every later import of the same URL throws 'Failed to fetch dynamically imported module' WITHOUT a network request: the new-session launch frozen in the loading hold forever (live-verified root cause of the stuck-loading arc; the rig has no tunnel so it never sees the class). retryImport parses the failing URL, re-imports it cache-busted (fresh module-map key) with backoff 1s→2s→4s→8s→16s (~31s of transport-storm ride-out; observed watchdog windows last minutes — a single immediate retry drowned in the storm's first beat); chained busts strip the previous query. Wired at all 8 dynamic-import sites. Transient-only: real errors rethrow. 7 contract tests (importer injected — the engine's import() never runs in the suite). Manifest: files + classification registered, drift gate PASS. * fix(app): bound dynamic import retries --------- Co-authored-by: amicode-ci <ci@amicode.local>
* fix: preserve interrupted compaction output * fix(app): localize compaction failure toast * fix(core): surface compaction runner errors * fix(engine): send final-step directive as user input * test(engine): cover manual compaction runner request * chore(app-bundle): refresh overlay manifest * fix(server): map compact runner failures --------- Co-authored-by: amicode-ci <ci@amicode.local>
Co-authored-by: amicode-ci <ci@amicode.local>
* fix: bound background session warming * test: cover prewarmer batch scheduling * refactor: share session prewarmer batching * test: cover multi-row prewarmer batches * fix: parallelize bulk lineage warming * fix: release warm slots after prefetch --------- Co-authored-by: amicode-ci <ci@amicode.local>
…leet-main # Conflicts: # packages/app-bundle/manifest.json
The 'does not walk past a secondary directory' test wrote AGENTS.md to path.dirname(tmpdirScoped()) === os.tmpdir(), an UNSCOPED file in the shared $TMPDIR root that (a) was never cleaned up and (b) poisoned every sibling test nesting tmp dirs under $TMPDIR — the loader walks up and finds it. Nest the secondary under its own scoped parent so the 'above' AGENTS.md lands in an auto-cleaned dir. findUp for a secondary dir stops at the dir itself (stop=dir), so no git root is needed. Fixes 2 instruction.test.ts failures.
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Advanced Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
jeonghun-jj-lee
pushed a commit
that referenced
this pull request
Sep 24, 2026
PR #1520 was squash-merged, which applied main's CONTENT but left main's 7 commits out of feature's ancestry — so GitHub kept showing 'feature 7 behind main'. This is a real merge commit (2 parents) that makes those commits true ancestors. Zero net content change: the merged tree is byte-identical to the squash result (6b9dd59); drift gate PASS (manifest ↔ overlay in sync).
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
3 participants
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.
Reconciles
feature/free-tier-fleetwithorigin/main(brings in main's 7 commits: session-reliability fixes, Open VSX CI, app-bundle overlay work), plus a real engine-test-hygiene fix surfaced while verifying.The merge
packages/app-bundle/manifest.json(both sides appended overlay entries). Resolved by regenerating it from the merged overlay tree viarefresh_manifest.mjs(hand-merging content hashes would be wrong) — drift gate PASS, 869 overlay files verified in sync. All actual source files merged cleanly.No regressions (the ask) — verified
/event→ 200)Engine unit suite — 11 pre-existing failures triaged; the real one fixed
The non-blocking engine:unit lane (#1239) showed 11 failures. None are reconcile regressions:
AGENTS.mdtopath.dirname(tmpdirScoped()) === $TMPDIR, an unscoped file that never got cleaned and poisoned sibling tests (the loader walks up and finds it). Fixed (nest the secondary under its own scoped dir;findUpfor a secondary stops at the dir itself, so no git root needed). Now 0 fail in isolation.OPENCODE_DISABLE_EXTERNAL_SKILLS=true(set by the Amicode terminal) leaked intobun test. In a clean env (as CI runs) → 17 pass / 0 fail.Clean-env full runs went 4 fail → 1 fail across identical invocations, confirming the residual is non-deterministic #1239 flakiness on the non-blocking lane — not a deterministic regression.