OpenClaw merged a large channel-account cleanup this morning in PR #114395, "refactor(channels): fix account resolution and setup isolation". The PR is framed as a refactor, but the user impact is operational: account selection should now agree across setup, runtime, and status for a long list of bundled channels.
The core problem was drift. Channel account listing, default selection, root-versus-named configuration, and credential replacement had grown into separate implementations. That made equivalent operations disagree. The PR specifically calls out Twitch inspecting the wrong selected account and credential replacement leaving a previous secret, file, or environment alternative in place.
What Changed
The patch extends OpenClaw's existing public-SDK account-list helper with opt-in merged-account resolution. It avoids adding a new public helper, callable, package subpath, configuration surface, or compatibility shim.
Owners now declare the account behaviors they actually support, including:
- Normalization rules.
- Real implicit or binding-discovered accounts.
- Empty-account behavior.
- Statically checked omitted properties.
- Additive nested configuration.
The setup patcher also clears only owner-selected conflicting credential fields before applying replacements. That matters because replacing a credential should not silently leave another credential source active beside it.
Channel Coverage
The migration touches a wide set of bundled integrations: ClickClack, Discord, Google Chat, iMessage, IRC, LINE, Mattermost, Nextcloud Talk, Nostr, QA, QQ Bot, Raft, SMS, Synology Chat, Telegram, Tlon, Twitch, Zalo, and Zalo User.
Several channel-specific contracts are intentionally preserved. Matrix stored-account discovery, WhatsApp authentication directories, Slack deep configuration, Signal transport configuration, LINE insertion-order and default semantics, and Telegram agent bindings stay with their owners.
That is the right shape for this kind of cleanup. Channel setup bugs often come from making a shared helper too broad. This PR does the opposite: it centralizes the repetitive account-list machinery while keeping sensitive owner rules explicit.
User Impact
For operators, the important outcome is consistency. Existing channel configuration and secret references keep their current semantics, but account selection no longer has to depend on which surface is looking at it. Setup, runtime, and status should agree.
The PR also says replacing supported credentials no longer accidentally retains conflicting alternatives, empty channel configuration does not invent accounts for channels that intentionally return none, and two dead channel exports were removed.
The size is notable: production account and setup code is reduced by 358 net lines across 28 production files, while regression tests cover sensitive discovery, disabled-account behavior, named-account inheritance, and binding selection.
The Verification Trail
The PR reports 934 passing tests across 53 focused channel and core files plus 17 Vitest project shards. It also lists lint, core and extension typechecks, plugin SDK API and surface checks, dependency checks, build, diff checks, and a source review of the full 40-file branch.
The labels also explain why this made the morning cut: P2, many channel labels, proof: sufficient, merge-risk: compatibility, and merge-risk: auth-provider.
Bottom Line
PR #114395 is a maintenance-heavy change with real product impact. OpenClaw channel accounts are one of those areas where invisible consistency is the feature. If setup, runtime, and status all resolve the same account, operators get fewer confusing channel failures and fewer stale credentials hiding in config.