test: add regression coverage for logout, provider methods, and tool-call merges

[cursor]

Automated test-coverage sweep. Adds 48 new tests (0 changes to production code) targeting recently-changed protocol surfaces that previously shipped without direct unit coverage.

Risky behavior now covered

Newly stabilized logout method (PR agentclientprotocol#1273)

logout was promoted from unstable_logout to the stable API without any accompanying tests. A silent break here would affect every client that upgrades.

• logout method-name constant and AGENT_METHOD_NAMES.logout value.
• ClientRequest::LogoutRequest(...).method() dispatch.
• LogoutRequest / LogoutResponse empty-object wire format and round-trip.
• LogoutResponse: Default contract (relied on by AgentResponse's #[serde(default)]).
• AgentAuthCapabilities default omits logout; .logout(LogoutCapabilities::new()) emits { "logout": {} }.
• Malformed auth.logout value is silently ignored via the DefaultOnError guard.
• Backwards compat: AgentCapabilities deserializes with auth absent.

Provider method rename (PR agentclientprotocol#1272)

PR agentclientprotocol#1272 renamed provider method types to singular (providers/set / providers/disable). Locks in the singular form so a future rename can't silently drift back and break already-shipped clients.

• PROVIDERS_SET_METHOD_NAME / PROVIDERS_DISABLE_METHOD_NAME constants.
• ClientRequest method dispatch for ListProvidersRequest, SetProviderRequest, DisableProviderRequest.

ToolCall merge semantics and wire-format guards

ToolCall / ToolCallUpdate drive the tool execution surface clients render live to end users, but the module previously had zero unit tests. A quiet regression would corrupt user-visible state.

• ToolCall::update applies all provided fields and leaves untouched fields alone (including the tricky raw_input/raw_output streaming case).
• Empty ToolCallUpdateFields leaves the call unchanged.
• Content and location collections are overwritten (per docs), not extended.
• TryFrom<ToolCallUpdate> for ToolCall errors with InvalidParams + the documented message when title is missing; uses defaults for every other field.
• ToolCall -> ToolCallUpdate -> ToolCall round-trips through TryFrom losslessly.
• Default ToolKind / ToolCallStatus are skipped on serialization (compact wire format, older-client compat); non-default values still round-trip.
• Unknown kind values decode to ToolKind::Other via #[serde(other)] (forward compat).
• Malformed entries in content / locations streams are skipped by VecSkipError rather than failing the whole update.

Test files added / updated

• src/v1/agent.rs: +10 logout tests, +2 provider-method tests.
• src/v2/agent.rs: same suite mirrored for v2.
• src/v1/tool_call.rs: new tests module with 11 tests.
• src/v2/tool_call.rs: same suite mirrored for v2.

Total: 316 tests (up from 268), all deterministic, all pass under cargo test --all-features. cargo clippy --all-features and cargo fmt --check also pass. The only prettier warning (AGENTS.md) predates this branch and is unrelated to these changes.

Why these tests materially reduce regression risk

• Wire-format lock-in on newly-stable methods. Once logout is out in released clients, any silent drift in the method name, request/response shape, or capability advertisement breaks every consumer. These tests turn each of those into a compile-time-caught failure.
• Guards against reintroducing the plural provider names that PR fix(unstable): Rename provider method types to singular agentclientprotocol/agent-client-protocol#1272 explicitly fixed.
• Semantic contract lock-in on ToolCall::update, which is invoked on every streaming tool update and whose behavior is only otherwise documented in comments. Overwrite-vs-extend and "don't clobber raw input on partial updates" are exactly the kind of subtle behaviors that regress unnoticed.
• Deserialization robustness. Tests exercise the DefaultOnError and VecSkipError serde guards on paths that receive untrusted input from a peer implementation, ensuring one malformed entry from a bad agent can't crash the client.

  

---

Summary by cubic

Adds 48 regression tests to lock down the stabilized logout method, singular provider methods, and ToolCall merge and wire-format behavior across v1 and v2. Also fixes AGENTS.md table formatting to satisfy Prettier; no production code changes.

• Test Coverage

  • logout: confirms constant and dispatch are "logout", empty-object request/response, default AgentAuthCapabilities, ignores malformed auth.logout, and tolerates missing auth.
  • Providers: enforces providers/list, providers/set, providers/disable constants and ClientRequest dispatch (singular form).
  • ToolCall: verifies update applies fields, overwrites collections, preserves absent raw_input/raw_output; TryFrom requires title and otherwise uses defaults; ToolCall <-> ToolCallUpdate round-trip; skips default kind/status on serialize; unknown kind → Other; skips malformed content/locations entries.

• Bug Fixes

  • AGENTS.md: formatted the commands table to pass Prettier.

^{Written for commit 54aa5b0. Summary will update on new commits.}

Note

Add regression test coverage for logout, auth capabilities, provider methods, and tool-call merges

• Adds tests in src/v1/agent.rs and src/v2/agent.rs covering logout method constants, LogoutRequest/LogoutResponse serialization round-trips, auth capability defaults, and provider method name constants (providers/list, providers/set, providers/disable) under the unstable_llm_providers feature flag.
• Adds tests in src/v1/tool_call.rs and src/v2/tool_call.rs covering ToolCall::update field application, collection overwrite behavior, TryFrom<ToolCallUpdate> error on missing title, serialization of default/non-default fields, unknown ToolKind fallback, and DefaultOnError skipping of invalid entries.

^{Macroscope summarized 54aa5b0.}

---

Note

Low Risk
Test-only and documentation formatting changes; no runtime or schema behavior is modified.

Overview
Adds 48 new unit tests across v1/v2 with no production code changes, locking in protocol behavior that recently shipped without direct coverage.

logout stabilization (PR agentclientprotocol#1273): Tests in agent.rs assert the "logout" method name and ClientRequest dispatch, empty-object serde for LogoutRequest/LogoutResponse, AgentAuthCapabilities default vs { "logout": {} }, malformed auth.logout ignored via DefaultOnError, and missing auth on AgentCapabilities.

Provider rename (PR agentclientprotocol#1272): Behind unstable_llm_providers, tests pin singular wire names providers/list, providers/set, and providers/disable plus matching ClientRequest method strings.

Tool calls: New tests modules in tool_call.rs cover ToolCall::update (partial fields, collection overwrite, preserving raw_input when absent), TryFrom<ToolCallUpdate> title requirement, round-trips, default kind/status omitted on serialize, unknown kind → Other, and skipping bad content/locations entries.

Docs: AGENTS.md command table is reformatted for Prettier alignment only.

^{Reviewed by Cursor Bugbot for commit 54aa5b0. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_bf8b00f7-9eb8-4839-b3c4-9d9d0329dd39)

[Copilot]

Pull request overview

This PR strengthens the protocol schema library’s regression safety by adding targeted unit tests (no production/runtime logic changes) that lock in recently shipped protocol behavior around logout, provider method naming, and ToolCall merge/serde semantics.

Changes:

• Add v1/v2 regression tests for stable logout wire format, method dispatch, and capability defaulting/robust deserialization.
• Add v1/v2 tests (behind unstable_llm_providers) pinning provider method names and ClientRequest dispatch strings to the singular forms.
• Add new ToolCall test modules in v1/v2 covering ToolCall::update, TryFrom<ToolCallUpdate>, default-field serialization skipping, unknown-kind fallback, and VecSkipError handling.
• Reformat the AGENTS.md commands table to render correctly (and satisfy Prettier).

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
src/v1/agent.rs	Adds logout + provider method-name/dispatch regression tests for v1.
src/v2/agent.rs	Mirrors the logout + provider test coverage for v2.
src/v1/tool_call.rs	Adds a tests module validating ToolCall merge semantics and wire-format behavior for v1.
src/v2/tool_call.rs	Mirrors ToolCall merge/serde regression coverage for v2.
AGENTS.md	Fixes markdown table formatting for the key commands section.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.