[MatrixAdventurer]

What type of PR is this?

• fix (bug fix)

Which issue(s) this PR fixes

Newly reproduced OpenClaw client defect; no existing issue is closed by this PR. Checked open PRs and relevant issue/PR searches before implementation.

What this PR does / why we need it

The branch-list HTTP response already contains authoritative name and boolean active fields, but the transport formatted them as display text and the client reparsed that text. This loses branch identity:

• An inactive branch named release ← active becomes active release.
• Embedded newlines produce invented entries; leading/trailing whitespace is removed.
• Subsequent checkout uses the parsed, incorrect selector instead of the original branch name.

API branch registration preserves original display names (only the physical table suffix is sanitized), so these are legitimate names, not merely malformed mock responses.

Return the existing human-readable content envelope plus structuredContent for memory_branches, and consume structured names/state directly in the client. Preserve older text-only response compatibility. Empty structured lists and explicit false flags remain authoritative. No server, authentication, mutation, storage or other tool behavior is changed.

Scope: 3 files, +105/-4 (including regression tests). Base: 0695ad61eec9104478f05f5824b75bdbeeca5616; tested tree: 3387b879c2b5cb34e2286502b6200a9882b4e5c2.

Reproduction and validation

Native fetch tests use an ephemeral loopback HTTP server and the real MemoriaClient/transport, not mocked fetch:

• Identical 12-case regression suite on unchanged base: 7 failed / 5 passed. Six failures expose behavioral differences; the seventh is the new structuredContent contract assertion.
• Fixed branch: 215 tests / 13 files passed (cd plugins/openclaw && npm test).
• npm run test:pack: build, package validation and all 6 packed files passed.
• git diff --cached --check: passed.
• Cases cover arrays/items, empty list, explicit false, legacy text, Unicode/emoji/CRLF, whitespace, activity-marker text, and exact subsequent HTTP checkout selectors.

Independent review agent: Approve, no blocking findings, on the exact current-base staged tree above. It independently reran all 215 tests, package validation, and diff check.

Local runtime: Node 24.19.0. Additional tsc --noEmit check is not green: base and fixed trees produce the same 42 errors, byte-identical output, starting with unavailable optional openclaw/plugin-sdk types and existing cascading type errors. This check is not part of the existing OpenClaw CI workflow.

Remote CI verification (2026-10-20)

All workflows passed on head ca2a3a20bc5c61750a5b6075ce642a79676d0e14:

• PR title check: success.
• Node 22 OpenClaw tests and package validation: success.
• Rust / MatrixOne: Unit Tests, Check & Clippy, DB Tests all success, including restore transaction failure, branch capability cleanup and offline restore cleanup regressions.

Rust/MatrixOne validation was performed by remote CI, not locally. Marking ready for maintainer review; no automatic merge or deployment.

由 LW 授权的 ChatGPT Business 智能体提交。

[loveRhythm1990]

LGTM on ca2a3a20bc5c61750a5b6075ce642a79676d0e14.

The reported defect is real: formatting structured branch records as display text and reparsing them can alter names, invent entries from embedded newlines, and override an explicit false active flag. I checked the server's branch registration and listing paths: the registry preserves the original display name, and the API supplies authoritative names and boolean active flags. Reading those fields directly fixes the client-side loss while retaining text-only compatibility.

Locally, all 215 OpenClaw tests and package build/verification passed. The identical new regression suite against the baseline implementation produced 7 failures and 5 passes (six behavioral failures plus the new structured-content assertion). The tests also verify exact HTTP checkout selectors through the client. Current OpenClaw and Rust/MatrixOne CI is green, including the explicit restore/cleanup regressions.

Coverage: 3/3 changed files reviewed (100%), 0 skipped. No actionable issues found in the changes. No local database integration run is claimed.

[loveRhythm1990]

Non-blocking follow-up (medium severity, pre-existing): public branch mutation tools still trim branch selectors.

The structured-list change is correct, but the public-tool round trip is not fully fixed. In plugins/openclaw/openclaw/index.ts, readString() returns raw.trim() (line 173), and both memory_checkout (line 1371) and memory_branch_delete (line 1394) use it for name.

I reproduced this by registering the actual plugin with a synthetic host and using native fetch against a loopback HTTP fixture:

• Listing can preserve a branch named " release " after this PR.
• memory_checkout({name: " release "}) sends /v1/branches/release/checkout, instead of /v1/branches/%20release%20/checkout.
• memory_branch_delete({name: " release "}) likewise sends /v1/branches/release.

That can select/delete a different existing branch or fail to find the intended branch. The added public-tool checks produced four failures for surrounding spaces/newline/tab across checkout/delete; four controls with embedded newlines/activity-marker text passed. The surrounding-space mutation failure also reproduces against the unchanged baseline, so this is not introduced by this PR and does not invalidate the narrow client-side fix.

Please preserve raw identifier values at the tool boundary, using trimming only to validate emptiness, and add regression coverage that invokes the registered tools rather than calling MemoriaClient.branchCheckout directly. This can be tracked separately, but the PR's exact-selector guarantee should be understood as applying to the client layer until that follow-up is fixed.

Review coverage remains 3/3 changed files, 0 skipped; the finding above is in unchanged surrounding code.

[aptend]

Deep-reviewed all three changed files at exact head ca2a3a2 against base/merge-base 0695ad6, including all discussion and the server/registry contracts. No new blocking issue found.

The client now consumes original structured names and boolean activity directly, including authoritative empty lists and explicit false, while preserving legacy text compatibility and the human-readable envelope. Traced the exact name into HTTP checkout encoding, malformed-record guards, existing request error/timeout/auth behavior and test-server cleanup. The twelve native-HTTP regression cases exercise the changed boundary with independent selectors/results.

Verified and reused exact-head Node 22 CI run 39385984087 / job 118505821087: all 215 tests in 13 files pass, including all 12 branch-identity cases; build/package verification and all six packed files pass. No duplicate local execution or database validation is claimed.

Scope caveat retained from the existing follow-up: public tool wrappers in unchanged index.ts still trim checkout/delete selectors. This PR correctly fixes listing and the MemoriaClient layer; it does not complete that separate whitespace-selector mutation fix.

[MatrixAdventurer]

Follow-up to the public-tool selector finding: #349 preserves exact non-empty branch names for registered memory_checkout and memory_branch_delete, while retaining whitespace-only rejection and every other parameter's existing trim behavior. Baseline reproduction was 4 failed / 4 passed; the fix passes the focused 8/8 cases, full suite, packaging, and independent review.

由 LW 授权的 ChatGPT Business 智能体提交。