docs: close out the 1.1.6 changelog heading - #70
Conversation
skip_cache shipped in 1.1.6 (absent at tag v1.1.5, present at v1.1.6) but CHANGELOG.md still files it under Unreleased. The release fan-out never touches this file — prepare_target.py leaves it alone as a historical record — so no release PR has closed it out and none will. Salvaged from #66, which closes the same heading out as 1.2.0. That version is wrong for this repo and #66 is otherwise obsolete: roe-main 3ae83b888 (#3747) removed the cross-repo version lockstep guard it existed to satisfy. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The hand-written response fixtures behind the connections and agents wrapper transport tests have drifted from the generated models. Three roe-main changes added read-only-but-required serializer fields that nothing downstream backfilled: #3758 dynamic_inputs, dynamic_input_test_disabled_reason #3837 credentials_configured #3701 BaseAgent.updated_at Connection.from_dict / BaseAgent.from_dict pop required keys with no default, so each addition turned into a KeyError. main has been red on these same four tests since #68 merged on 2026-07-29 (run 30493974267); this PR did not introduce the failure, it only changed which missing key raises first, because credentials_configured is popped before dynamic_inputs. Fixing it in this branch rather than separately so #69 merges green — publish.yml is `on: push tags v*` with no test step and no needs: on CI, so a red merge would tag v1.1.7 and ship an untested tree to PyPI. Only the response fixtures change. The expected request-body literal in test_connection_create_... is deliberately untouched: these are read-only fields the client never sends. tests/ is outside the python target's diff_paths and outside check-codegen-drift, so the fan-out will not clobber this. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Greptile SummaryThis update moves the existing The checked release metadata and tag context match the new heading, the feature notes remain in place, and the resulting changelog structure is valid. Confidence Score: 5/5
What T-Rex did
Reviews (1): Last reviewed commit: "docs: close out the 1.1.6 changelog head..." | Re-trigger Greptile |
|
Added the stale-fixture commit here as well (cherry-pick of Context: The commit is byte-identical on both branches, so either merge order is fine — whichever lands second is a no-op for those two files. Merging this one also un-reds |
Why
CHANGELOG.mdstill files theskip_cachefeature under## Unreleased, but it shipped in 1.1.6 — verified against the tags:git show v1.1.5:src/roe/api/agents.py | grep -c skip_cache→0, and the same command atv1.1.6→17.This is not a one-off oversight. The release fan-out never touches this file by design —
roe-sdk/src/roe_sdk/prepare_target.py: "CHANGELOG.md is intentionally not touched (historical record)". So no release PR has ever closed the heading out, and none will. Left alone, 1.1.7 ships with an empty changelog too.The only open PR that touches
CHANGELOG.mdis #66, which closes the heading out as1.2.0— the wrong version (1.2.0 was a roe-golang-only bump, and this repo's live line is 1.1.x continuing to 1.1.7 in #69). #66 should be closed rather than merged: its reason to exist was deleted upstream by roe-main 3ae83b888 ("Version SDK release targets independently", #3747), which removed the cross-repo version lockstep guard. This PR salvages the one genuinely useful line from it, at the correct version.What
One line:
## Unreleased→## 1.1.6.No version file changes, so
tag-on-release-merge.ymldoes not fire and nothing publishes.The same fix is up for roe-typescript, whose
CHANGELOG.mdhas the identical stale heading over itsskipCacheentry.🤖 Generated with Claude Code