Repository navigation
fix: 0.12.1 batch 39 - Bolin queue picks and Slack-bound ledger fix - #32572
Conversation
…32368) A template whose spec could not be read during the first agent scan in a gateway process is missing from the snapshot while its file is on disk. Only the main-chat turn rescanned on that miss, so a template chat create and a template agent switch answered 409 agent_choice_unavailable until something else rescanned or the gateway restarted. Both now await rescan_if_incomplete() before refusing: one off-loop scan while the snapshot is incomplete, then one more resolve. A complete snapshot is trusted and costs no scan. Closes #27878 (cherry picked from commit c460a20)
…32309) A session_create caller with no execution record was read as the crewmate its slot's agent name points at, so it inherited that crewmate's model pins and was treated as its conductor. A slot name is metadata, not an attested identity. _dispatching_crewmate now answers "no crewmate" for such a caller, so every reader gets the plain behaviour, and the per-caller conductor_record_missing guard, which that answer makes unreachable, is removed. Closes #23834 (cherry picked from commit 45775d8) Original author: Bolin Chen (#32309). One doc hunk resolved by hand in docs/system-specs/modules/session-control.md: took #32309's sentence and kept this branch's "mate" wording (main's "Bot" rename is main-only).
#32386) spawn_sub_agents stops waiting for a member once the spawn gate defers it, but the gateway's waiting_children record for the blocked parent still listed that member, so the parent's lane slot was not granted back until the tool's max_wait ran out (up to two hours). The tool now names the members it settled as deferral-held once, through an internal release route only the attested run itself may call. The gateway keeps those ids in a per-run released set and resumes the parent once every other awaited child has ended, without editing the wait record; no later wait of the run awaits them again. Builds on #26052 by Pearcekieser. Co-authored-by: Pearce Kieser <5055971+Pearcekieser@users.noreply.github.com> (cherry picked from commit 8c4f4a4) Original author: Bolin Chen (#32386). One hunk resolved by hand in test/test_dashboard_server_composition_contract_reach.py: the MCP route table pin is recomputed for this branch (332 -> 333 rows), since the branch has fewer routes than main.
…32445) An IAM Identity Center sign-in asked only the control plane picked from the Identity Center region for the account's Kiro profiles. That region does not say where the profile is served, so a us-east-1 instance whose profile lives on the eu-central-1 control plane got zero profiles and every sign-in ended with "no available Kiro profiles". Ask every known control-plane host, the login region's own first, and keep the first answer that lists a profile. A host that refuses is logged and skipped; the login fails only when every host refused or none lists a profile. Closes #28904 (cherry picked from commit 90d19e3)
A spawn_sub_agents child whose last status read failed (a timeout, a 404 or a 503 from a loaded gateway) was reported as status error and counted as errored, while the child kept running. The caller moved on, and the child's completion event later landed in a parent turn that had been told it failed. The gateway sends error only on a finished row, so an error without done is a failed read. The collect step now lists such a child under still_running as unread, with its id, and keeps error entries for runs that finished badly. The envelope's waited_secs reports the time this call actually waited, since a failed read ends the wait early. The deferred-member set that only fed the old hint is removed. Closes #25425 (cherry picked from commit 51c2aaf)
…32324) A task-runner step, a workflow agent step and a heartbeat turn sent their first prompt about a second after the first MCP server reported, so a slow server could miss that turn and the run went on without its tools, with nobody there to reconnect it. On a new session these runs now wait, up to 30 seconds, for every server the agent spec declares, the same bounded wait webhook and cron runs already use. A task-runner step names a declared server that never reported in the MCP notice it already sends, and a heartbeat names every not-ready declared server at the top of its result. Sub-agents and the tool-less calls are left out; the spec says why. Closes #21700 (cherry picked from commit f868145) Original author: Bolin Chen (#32324). One import hunk resolved by hand in src/kiro_crew/workflows/agent_exec.py: kept this branch's imports and added the mcp_problems import; main's acp_backends import block belongs to per-step backend work this branch does not carry.
…ht (#32317) On a server with caller identity, a stub's detach, a warm-pool claim's eviction and the release sentinel's grant arm each released the caller's upstream subscription while another stub's subscribe for the same URI as that caller was still in flight. The server then granted that subscribe and dropped the caller's one subscription, leaving the new subscriber routed to nothing. All three now decide the release through _hold_identity_release, as a final unsubscribe already does: the in-flight subscribe inherits the release and its verdict settles it. The sentinel arm decides per caller, so a stub of another caller routed for the URI no longer keeps the caller's unwanted grant alive; a held release with nobody routed orphan-marks the URI so its updates are not charged as a hazard. Reported by javenciu in #21433. (cherry picked from commit 21597a3)
…nits record (#32490) A nudge cycle reads a Slack-bound loop's work ledger by the loop's binding key (slack:<ts>) with no live session id, so canonical_slot kept that raw spelling. Both writers of a channel conversation head the unit with the filename fold (slack_<ts>), so the read found only units an older build headed with the raw key and injected their stale snapshot every cycle. A messaging-channel key now falls back to its transcript_stem fold, with the raw spelling still joined as the alias, and alias units are filtered by the alias's own exclusion list as well. (cherry picked from commit cc732f3)
UX Review (Opus 5.5) — ⏭️ skippedRevision |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS This PR only cherry-picks eight fixes already merged on Notes
[DESIGN-REVIEWED] 875b33e Description read[DESCRIPTION-READ] 283ed13c615ca53d96628b8f7fbbfa84fbaa9992db15250d28fdbffbf8ee57d7 That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
First Principles Review (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS Each item is a backport of a fix already merged on What this change shipsInventory (8 items) — 8 justifiedIntent: give 0.12.x users eight bug fixes that already shipped on
[FIRST-PRINCIPLES-REVIEWED] 875b33e Description read[DESCRIPTION-READ] 816f097d965fef39fad9af93570ac20d08f47d56dc06b7d89956f0703641adaf That sha256 names the PR description this verdict read. Recompute it with |
GPT 6.1 Review — ✅ no blocking findingsGPT 6.1 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
Opus 5.5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
Problem / Motivation
Goal: Bring eight fixes that merged on
mainafterv0.12.0(seven of Ray's 0.12.1 picks from Bolin's queue, plus the Slack-reported #32490) ontorelease/0.12.0, one commit each, in a PR of its own.Why it matters
session_createreads a crewmate from a slot name that has no execution record.spawn_sub_agentsreports a slow-starting child as failed while it keeps running, and its later completion kills a busy cron parent.Not a goal
main; nothing is changed only here.CHANGELOG.mdchange. Cuttingv0.12.1-rc.Nand the bare tag follow docs/build/release.md and stay with the maintainers.What changed (motivation → approach → change)
Each main squash commit was cherry-picked with
-xontorelease/0.12.0@44e6f443e4, in merge order onmain, one commit each.c460a2063b6d6fde399545775d89872ccbf0c6f98c4f4a42be229de5ade090d19e354989c5fe881a51c2aaf1ec75a01f8718f8681455f6dd801b646121597a32543d887fba03cc732f39dc875b33eed7Hand-resolved hunks. No prerequisite was pulled in: each conflict is context from a
main-only change next to the pick's own lines.docs/system-specs/modules/session-control.mdtest/test_dashboard_server_composition_contract_reach.py2c2e11d2…(main has more routes)src/kiro_crew/workflows/agent_exec.pymcp_problemsimport; main'sacp_backendsimport block belongs to per-step backend work not on this branch.wait_for_declared_mcpandUNATTENDED_MCP_WAIT_SECSalready exist hereEvery other file has added and removed lines identical to its main commit. Each hand-resolved commit is authored with the public identity the push guard requires and names its original author in its message; clean picks keep their original author.
builder-security checked: no bind, listener, route exposure, auth mechanism or storage grant changes. #32445 changes which control plane the IdC profile lookup asks; #32386 adds one internal release route that only the attested run itself may call.
Backwards compatibility
Compatible: each change is the one already merged on
main, and its compatibility note applies unchanged.Removes nothing: no setting, key, route or caller is removed.
Tests
Proof: not provable -- other: a backport of main commits whose proofs ran on their own PRs
release/*, so the related set was run here: every touched test file plus every test file that imports a touched module, 588 files,-n 4 --dist loadgroup. 22586 passed, 11 skipped. 4 failed, none from these picks:test_harness_parity.py::test_is_kiro_cli_is_positiveandtest_auto_improvement_members.py::test_assignments_build_real_context_from_only_their_private_store: failures this branch already carries.test_gateway_secret_senders_prove_peer.py::TestArtifactSendsThroughThePeerProof::test_list_goes_through_the_proof: fails the same way on a cleanrelease/0.12.0@44e6f443e4.test_home_pin_survives_monkeypatch_undo.py::test_the_guard_accepts_a_scratch_home: passes when its file runs alone on this branch (29 passed); an order effect under xdist.Manual verification
N/A - each fix was verified on its own main PR. The 0.12.1 RC soak is the end-to-end check.
Screenshots / video
N/A - no frontend file changes.
Related Issues
Backport to the 0.12 line for 0.12.1 of #32368, #32309, #32386, #32445, #32447, #32324, #32317 and #32490. Source issues: #27878, #23834, #26051, #28904, #25425, #21700, #21433.
Sibling sweep
Siblings: none -- each source PR carried its own sibling sweep; this PR only carries those fixes.
Pattern harvest
Not generalizable: this PR only carries fixes already on
main. The source PRs have their own harvest.Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)Contribution License Agreement
Contributions are accepted under the project's Apache-2.0 license; see the Licensing section of CONTRIBUTING.md.