Repository navigation
docs: backfill CHANGELOG.md for releases since the version sync - #105
Conversation
Adds a section for every tagged release after 1.0.802 that had none, generated from each release's openapi/openapi.yml and openapi/wrappers.yml diff by roe-main's roe_sdk.changelog (the same code the release fan-out now uses), then reviewed by hand. Existing hand-written sections are unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
[Low impact] The PR changes the release history readers see, not SDK behavior. This documentation-only PR appears safe to merge, with three non-blocking corrections to release history.
|
|
|
||
| ### Removed | ||
|
|
||
| - **Breaking.** Removed the `NullEnum` model from `roe._generated.models`; update any imports or type references that still use `NullEnum`. |
There was a problem hiding this comment.
NullEnum never existed in Python
The Python SDK never exported NullEnum. In 1.1.12, BaseAgent.skill_generation_state already uses None | SkillGenerationStateEnum. The 1.1.13 change removes a schema entry without changing the generated Python code. Remove the claimed Python addition and breaking removal from these two sections so readers are not sent to fix imports that never existed.
| - Added the `client.agents` API with methods to manage agents and run them: `client.agents.create(...)`, `client.agents.list(...)`, `client.agents.retrieve(...)`, `client.agents.update(...)`, `client.agents.delete(...)`, `client.agents.duplicate(...)`, `client.agents.run(...)`, `client.agents.run_many(...)`, `client.agents.run_sync(...)`, `client.agents.run_version(...)`, and `client.agents.run_version_sync(...)`. | ||
| - Added `client.agents.jobs` methods to manage agent runs and results: `client.agents.jobs.cancel(...)`, `client.agents.jobs.cancel_all(...)`, `client.agents.jobs.delete_data(...)`, `client.agents.jobs.download_reference(...)`, `client.agents.jobs.retrieve_result(...)`, `client.agents.jobs.retrieve_result_many(...)`, `client.agents.jobs.retrieve_status(...)`, and `client.agents.jobs.retrieve_status_many(...)`. | ||
| - Added the `client.agents.versions` API for agent version management with `client.agents.versions.create(...)`, `client.agents.versions.list(...)`, `client.agents.versions.retrieve(...)`, `client.agents.versions.retrieve_current(...)`, `client.agents.versions.update(...)`, and `client.agents.versions.delete(...)`. | ||
| - Added the `client.policies` API with `client.policies.create(...)`, `client.policies.list(...)`, `client.policies.retrieve(...)`, `client.policies.update(...)`, and `client.policies.delete(...)`, plus policy version methods `client.policies.versions.create(...)`, `client.policies.versions.list(...)`, and `client.policies.versions.retrieve(...)`. | ||
| - Added `client.users.me(...)` to retrieve the current user. |
There was a problem hiding this comment.
The APIs listed under 1.1.1 already existed in 1.0.802. That release already wired up AgentsAPI, PoliciesAPI, and UsersAPI, including their listed methods. 1.1.1 adds these definitions to openapi/wrappers.yml, not to the public client. Describe that contract change instead of presenting the existing APIs as new, which gives readers the wrong minimum version.
|
|
||
| ### Added | ||
|
|
||
| - Added support for resending an agent job webhook with `client.agents.jobs.resend_webhook(...)`, and the corresponding raw operation `agents_jobs_webhook_resend_create` for `POST /v1/agents/jobs/{job_id}/webhook/resend/`. |
There was a problem hiding this comment.
Webhook method appears too early
client.agents.jobs.resend_webhook(...) is not available in 1.1.8. That release adds the raw operation and models, but the public method first appears in 1.1.9, already accepting webhook_id. Keep the raw-operation claim here and move the public-method introduction to 1.1.9; otherwise the documented call fails for readers using 1.1.8.
Summary
Backfills
CHANGELOG.mdfor every tagged release since the 1.0.802 version sync that had no section:1.0.803, 1.1.0–1.1.5, 1.1.7–1.1.23 (24 sections).
The existing hand-written sections are unchanged and stay in place. New sections are inserted in version order.
How it was generated
openapi/openapi.ymlandopenapi/wrappers.ymlbetween consecutive release tags.roe_sdk.changelogfrom roe-main, the same code the SDK release fan-out uses since roe-ai/roe-main#4719. From now on, each release PR writes its own entry.Review
I checked every bullet against the actual SDK code:
client.…method named exists in the wrapper contract at some release tag.I corrected 4 bullets where OpenAI credited a raw-operation query parameter to a client method. For example,
client.agents.list(...)was said to accept the newagents_listfilters, but it still takes onlypageandpage_size. Those bullets now name the raw operation and say the wrapper doesn't expose the filter yet.Known limitation: sections come from spec and wrapper changes only. Changes to hand-written SDK code that the spec doesn't capture, such as retry behaviour, aren't listed.
Test Plan
client.…references resolve to real methods.Merging this won't re-tag or re-publish anything. It doesn't change the version, and
tag-on-release-mergeskips versions that are already tagged.1.1.24 (open PR #104) isn't included. Once it's merged and tagged, re-run the backfill to add its section, or write it by hand.
🤖 Generated with Claude Code