Repository navigation
fix: backport the edition MCP tools ref and KAS OAuth pointers to 0.12.1 (batch 33) - #32229
Conversation
UX Review (Opus 5.5) — ⏭️ skippedRevision |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing blocks: both picks match their main commits. Every Notes
[DESIGN-REVIEWED] ce6eac6 Description read[DESCRIPTION-READ] b8e8983fa290e7691cb0fc36e39a16050d79d1d6f860baa5bbb122d5b935c0b5 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 backport carries two fixes already merged on Notes
What this change shipsInventory (6 items) — 5 justifiedIntent: bring two user-reported MCP fixes from
[FIRST-PRINCIPLES-REVIEWED] ce6eac6 Description read[DESCRIPTION-READ] 24f8f0753c705f63915c091c07efcab0668d9fc958787b3997f29cd934166a0b That sha256 names the PR description this verdict read. Recompute it with |
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: |
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: |
…ce (#31611) An edition server that extra_mcp_servers() starts supplying after the first install was seeded into mcpServers on rebuild, but its @name ref was only ever added to tools on a fresh install, so the agent mounted the server and saw none of its tools. The rebuild now adds the ref to tools (never allowedTools) once, on the first rebuild that mounts the server, and records each name in edition_mcp_refs_offered.json after the spec is written, so a ref the user removes stays removed. Gated-off and opt_in servers gain nothing and stay unrecorded. Each added ref writes one mcp_tools_added SEL record. (cherry picked from commit 8b8cd96)
…em (#31671) A remote MCP server with a pre-registered OAuth client could not sign in on KAS, because the engine's customAgents schema dropped the oauth block and the registry marker. Newer engines announce mcpServerOAuth and mcpRegistryPointers in initialize. When they do, the projection sends oauth (clientId, redirectUri, clientMetadataUrl, oauthScopes; never clientSecret) and sends a user server's registry marker as a pointer that keeps its command and args, never its env. Crew's own servers are never pointers: they stay stdio so the session-level hoist still stamps their session identity. An engine that announces neither gets the same request as before. (cherry picked from commit b247a22) Release backport, hand-resolved. Each conflict kept release's side and added only this change's own lines: kwarg/parameter additions in acp/harness/kas.py, acp/kas_agents.py and two test files, the keeps_mcp_oauth import in dashboard/chat_turn/turn_spec_hooks.py, and the _agent_capabilities field in acp/runtime_session_setup.py's docstring. acp/harness/kas_cloud.py is deleted on this branch, so its one-line hunk was dropped. Every other file applied unchanged. Original author: Ray Xu (#31671).
3e43c95 to
ce6eac6
Compare
Problem / Motivation
Goal: Bring two user-reported MCP fixes that merged on
mainafterv0.12.0ontorelease/0.12.0, one commit per fix, in a PR of their own so they can be reviewed and reverted apart from the low-risk batches.Why it matters
@<name>ref never reaches the agent'stools, so the agent cannot see its tools. Only fresh installs get it.oauthblock, so a pre-registered OAuth client can never sign in.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.mcpServerOAuthormcpRegistryPointers: with neither announced, every path below behaves as on 0.12.0.What changed (motivation → approach → change)
Each main squash commit was cherry-picked with
-xontorelease/0.12.0@344cc7ca44, in merge order, one commit each.8b8cd96a971a62f3a32fb247a22121ce6eac60c1Hand-resolved hunks in #31671. Each conflict was traced to the
maincommits that added its context: #30961 (prompt_unread_out, in #32222), #30874 (with_switched_off_tools), #31562 (in #32174), #29543 and #30108 (imports), and #25347 (Kiro Cloud Sessions, which createdkas_cloud.py). None was pulled in as a prerequisite: each is its own fix or feature, and #25347 alone is 88 files. Each hunk keeps release's side and adds only #31671's own lines.acp/harness/kas.pyprompt_unread_out=prompt_unread(from #30961, now on release) and addedengine_features=engine_featuresafter it, as onmainacp/kas_agents.pyprompt_unread_outparameter and addedengine_featuresafter it, as onmainacp/runtime_session_setup.pymainword for word (_agent_capabilitiesadded,_child_env_nameskept)dashboard/chat_turn/turn_spec_hooks.pykeeps_mcp_oauthimporttest/test_kas_agents_mcp_servers.py,test/test_member_dispatch_mount.pyprompt_unread_out=Noneand addedengine_features=frozenset()in five fake signatures, as onmainacp/harness/kas_cloud.pyengine_featuresparameter there has nowhere to goEvery other file has added and removed lines identical to the main commit, including the new
acp/kas_engine_features.pyand its tests. #31671's commit is authored with the public identity the push guard requires and names its original author in its message; #31611 keeps its original author.builder-security checked: no pick changes a bind, a listener, a route or an ingress rule. #31671 changes what reaches KAS for a remote MCP server's sign-in: its
oauthblock, narrowed bycarry_oauthto the keys the engine accepts over ACP plusoauthScopes, sent only to an engine that announcesmcpServerOAuth. AclientSecretis never sent (one log line names the field, never its value). Otherwise Crew keeps withholding the block as on 0.12.0.Backwards compatibility
Compatible: each change is the one already merged on
main, and each PR's own compatibility note applies unchanged. Until the engine announces the flags, #31671 is inert.Removes nothing: no pick hides, deletes or renames anything 0.12.0 accepts.
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, 224 files, one pytest process per batch of 20,-n 0. 12819 passed. 11 failed, none from these picks: this branch's known copilot,test_is_kiro_cli_is_positiveandtest_folder_steering_provider_dedup.pyfailures, plustest_acp_agent_env.py::test_every_backend_launch_carries_agent_env[copilot]andtest_acp_tool_gate.py::test_every_enforced_harness_reaches_the_spawn_preflight, which fail the same way on a cleanrelease/0.12.0checkout.Manual verification
N/A - every pick was verified on its own main PR. The 0.12.1 RC soak is the end-to-end check.
Screenshots / video
Why no screenshot: backend-only change.
Related Issues
Backport to the 0.12 line for 0.12.1 of #31611 and #31671. Source issue: #27711.
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. Each source PR has its 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.