Repository navigation
fix: v0.12.2 hotfix batch 4 (four more of our fixes, adapted to release) - #34746
Conversation
…3818) An edition's cloud_instance_policy names managed policies that every EC2 crew's instance role gets, but nothing checked that the launch account holds them. In an account without one, CloudFormation's attach failed with IAM NoSuchEntity and the stack rolled back, leaving the user only the raw 404 from cloud launch, setup or the dashboard. ec2.deploy asks IAM for each such ARN before the source upload and the network discovery, and refuses the launch with a message naming the policy when IAM's error code is NoSuchEntity. The code is read from the aws CLI's own error prefix, never from the text that quotes the ARN. A read that fails another way is logged and the launch goes on, since the attach still needs the policy. The --host-policies addendum grants iam:GetPolicy on exactly those ARNs while the document fits IAM's 6,144-character cap. Without room it is the attach statement alone, and the CLI says so on stderr. The cloud and platform-context specs no longer say the policy exists in every account. (cherry picked from commit 6a1ea99)
…33644) Agent welcome: a new chat on an agent whose spec sets welcomeMessage showed nothing until the first send, and the row the first turn wrote then folded away with the working steps under the default collapse-all setting. The slot create, agent switch and workspace and project switches read the answering agent's greeting, the slot summary ships it as welcome_message while the chat has no row, and the welcome page draws it under its heading. The cold-start row carries meta.kind agent_welcome, and TurnBlock keeps that kind out of the fold. The greeting read runs after the switch locks are released, so a newer switch can commit during it. Each switch therefore reads its answer from the slot once that read returns, and a slot replaced under the name meanwhile answers the 404 a missing slot gets: the acting tab writes the answer into its store and must not bring back a superseded binding. The create's read runs after its block too, so a create whose chat is closed or replaced during it answers the same 404: the browser adds the create's answer to its list and drops the close's tombstone for that key, so a success would bring back a closed chat. Lazy chunks: the same fix shipped in #33687 (a failed chunk for the unrestored-tabs notice, the crew window or the composer chips stays in its place, and Try Again imports it again), so this commit keeps main's version of it and carries none of those hunks. Fixes #33593 Backport note: slot_projection.py imports only welcome_preview; release has no preparing_caption (main-only warm catch-up caption), so that import is not added. Backport note: slot_session.py docstring Writes line keeps release's autocompact_pct (main lists backend there) and adds _welcome_preview. Backport note: TurnBlock.tsx adds only the agent_welcome kind; release has no crew_paused_notice, so that kind and its comment sentence are not added. Backport note: index.chat.ts adds only welcome_message; preparing_text and preparing_lang do not exist on release and are not added. Backport note: ChatPage.views.test.tsx adds only the welcome greeting test; the warm catch-up caption test covers main-only code and is left out. (cherry picked from commit 208d6b2)
…fixes (#33510) Backport of the parts of #33510 that release/0.12.0 can carry: - tools allowlist: mcp.json stubs and KAS includeMcpJson follow the agent's tools, so a narrow agent no longer gets every mcp.json tool. A routed mcp.json server whose env holds a secret:// reference that tools leaves out gets a refusing stub instead, which starts nothing and offers no tools, so the host's own mcp.json merge cannot start it with the literal reference. - useLegacyMcpJson: the rewriter reads kiro-cli's alias through the reader the KAS projection uses (mcp_cleanup.include_mcp_json), so an agent with useLegacyMcpJson: false gets no injected mcp.json stubs. - glibc floor: below the engine's glibc 2.28 (Amazon Linux 2) chat, dashboard and doctor name the cause and a newer image, not a retry. - file:// prompts: the spec reference states kiro-cli v3 refuses an absolute or ~ file:// prompt and why Kiro Crew keeps accepting it. Both rewriter changes share one move of the rewrite fingerprint schema to 14, so a kept overlay holding stale injected stubs is rebuilt. Fixes #32973 Fixes #32923 Fixes #32896 (cherry picked from commit d4f82fb) Backport note: left out the trustedAgents on KAS fix. Release has no canonical_tools_settings, trusted_agents_withheld or per-session governance (may_skip_gate_for_session, session_profiles_govern), so the governance_app and governance_scope plumbing and its test and governance.md hunks are not carried. Backport note: left out the hook-covered grants fix. Its only answerer is the batch-guard-grant gate row and batch_guard._auto_approved, which are not on release, so CREW_ANSWERED_HOOK_HELD_KEY, hook_held_grants, spec_hook_held, pre_tool_hooks_fire and their tests and docs are not carried. Backport note: dropped the app files (apps/agent_turn_sdk.py, parked_queue owning_app, auto_improvement runners and its test) as out of scope; src/kiro_crew/acp/harness/kas_cloud.py, test_kas_tools_settings_aliases.py and test_app_agent_turn_sdk.py stay deleted as on release. Backport note: kas_agents_registered.py (a pure move of the projected_* and switched_* readers) is not carried; they stay in kas_agents.py. Backport note: kas_agents.py keeps release's imports and to_client_custom_agent body, with only the includeMcpJson-true-stays-off -the-wire change and the resolve_prompt docstring applied. Backport note: agent-spec-fields.md and security.md take release's text plus the includeMcpJson, prompt and secret:// refusing stub hunks; the hooks row and the hook-held security paragraph are not carried.
…#33820) A spec file in ~/.kiro/agents that the Custom agents list cannot show still made its name taken. The list keeps only files that read and one row per declared name, while a create counts the stem of every file, readable or not, so a create was refused 409 name_taken with no row to open, rename or delete, and the delete could not reach the file. GET /api/agents/templates now returns `unlisted`: each hidden file with its reason (it does not read, or it declares a name another file holds) and whether a delete can reach it. The tab shows them under Files not shown, with Delete file or a rename-by-hand hint; a same-name copy's confirm says the listed file and the agent of that name stay. A refused create names the file that holds the name, including the stem of a listed file whose row carries another name. DELETE reaches an unreadable spec by its exact stem under the same reference guard and a locked re-check. Deleting a hidden file whose declared name a listed file still declares keeps that name's sidecar entry and saved capabilities, and its reference and capabilities guards count the file's stem alone, since the name stays with the listed file. A file that holds nothing once the hidden file goes does not keep the name: a symlink that reads the hidden file, or a markdown spec while a configured host loads JSON specs alone. Beside such a listed file the tab offers no delete of the copy, since its confirm says the listed file stays. When agent.spec_pins picks the hidden file for that name, by naming it or a symlink that reads it, every session of the name runs it, so the tab offers no delete and the guards count the name too. A name a listed row carries keeps the plain refusal. A spec whose row cannot be built no longer breaks creates and deletes of other names, and neither one nested past the JSON parser's depth nor a symlink loop fails the tab or its own delete. Discovery logs a WARNING when it drops a plain file whose name another file declares, with the names escaped. Fixes #33664 Backport note: docs/feature-map/README.md and learn-cron-dashboard.md conflicted; kept release wording (mates) and re-applied the fix sentences by hand. Backport note: release has no agent.spec_pins (agent_spec_pins, agent_spec_pin_ops, declared_spec_claim are main-only); dropped pinned_here and every pin check from agent_templates_unlisted.py and agent_templates.py, the pin paragraph of crew-mode.md and the three spec-pin tests of test_agent_templates_unlisted.py. Backport note: AgentTemplatesTab.unlisted.test.tsx expects "a Bot" (release catalog wording) instead of "a bot". (cherry picked from commit 29a09dc)
The release branch has no black baseline row for the file the #33820 backport adds, so the black gate flags it. Format it. Backport note: assembly-only change, no behaviour change.
Sensitive Change ReviewSensitive files changed:
How to clear this check:
Only the author's comment counts this way. A reviewer can instead |
UX Review (Opus 5.5) — ✅ PASS (a BLOCK with no Trigger and no Outcome does not block)UX-level review of UX-Verdict: PASS cannot evaluate: missing screenshots of the new "Files not shown" section on the custom agents tab and the agent greeting on the chat welcome screen. The PR attaches no image, so the blind read was not performed. Evidence gaps
Notes
[UX-REVIEWED] cf5e0fc Description read[DESCRIPTION-READ] 8b0131318817a6b3ab2ed16768c0a2664cac49944b12aa82197e68a96a25278c That sha256 names everything this verdict read: the PR description, plus the 3 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 |
Design Review (Opus 5.5) — ✅ PASS (a BLOCK with no Trigger and no Outcome does not block)Design-level review of Design-Verdict: PASS cannot evaluate: missing a screenshot of the Agents tab's new "Files not shown" panel and of the agent greeting on WelcomeView. The description says "no visual change", but the diff adds both surfaces. BlockersA new UI surface ships without rendered evidence of this revision Notes
[DESIGN-REVIEWED] cf5e0fc Description read[DESCRIPTION-READ] 82f0e4ddcef9da9b38c1a2a11c3ded80967168bb6504ee252f575514c8312f32 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 This backports four already-merged main fixes. Each item removes a defect the Goal names, and no item makes a new default or product-shape decision. What this change shipsInventory (9 items) — 9 justifiedIntent: give 0.12 users four bug fixes from main without waiting for 0.13. This is a FIX.
[FIRST-PRINCIPLES-REVIEWED] cf5e0fc Description read[DESCRIPTION-READ] ff0c9f005bfeac3c1316ddc4953cb6e5adcd116077e029e42a2c9f3fe036897f 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: Ship four more of our own fixes from main to 0.12 users in 0.12.2, on
release/0.12.0.These four fix bugs that are in release code. They are larger than the usual pick, so each is adapted to release's code and leaves out what needs code only on main; every adaptation is a
Backport note:line in its commit.Why it matters
0.12 users hit each of these today: a cloud launch that fails with a raw IAM error, a chat greeting that appears only at the first send, KAS settings that drop mcp.json tools, and agent spec files the custom agents tab hides but still reserves.
Not a goal
git cherry-pick -xof the main merge commit; a hand-port keeps release's own code and says what it left out in aBackport note:line.What changed (motivation → approach → change)
The 8 picks, in main merge order, on
release/0.12.0atf0994b5e47:6a1ea993df208d6b2b89d4f82fbaf629a09dcca2Release-only commit: one black reflow of
agent_templates_unlisted.py(backport assembly).Hand-ports with a
Backport note::Each commit's body lists what it adapted and what it left out.
Backwards compatibility
No behaviour change beyond each listed fix. No config key, CLI flag or API is removed or renamed.
Tests
flake8,isort,mypy --platform linux, black, comment-history and brand: clean. Websitenpm run build,tsc -p tsconfig.app.jsonand the touched vitest files: pass.Proof: not provable -- other: a cherry-pick batch of already-merged main fixes, each proven in its own main PR
Manual verification
N/A: each fix was verified in its own main PR; the whole 0.12.2 line is smoke-tested again on the gate build.
Screenshots / video
N/A: no visual change.
Related Issues
0.12.2 hotfix, batch 4: #33818, #33644, #33510, #33820.
Sibling sweep
Siblings: none -- a backport batch; each pick's own PR carried its sweep
Pattern harvest
Not generalizable: a release backport batch, not a new defect fix.
Checklist
cherry-pick -xof a main commit with a Conventional Commits titleContribution License Agreement
Contributions are accepted under the project's Apache-2.0 license; see the Licensing section of CONTRIBUTING.md.