claude-code/docs/g007-plugin-mcp-verification-map.md

G007 Plugin/MCP Lifecycle Verification Map

Goal: G007-plugin-mcp — Stream 5 plugin/MCP lifecycle maturity from ROADMAP Phase 5.

Scope: worker-2 follow-up map for W4 mock integration and regression verification. This file intentionally does not mutate leader-owned .omx/ultragoal state.

Covered ROADMAP / CC2 anchors

• ROADMAP.md:55-57 — Current pain point §6: plugin/MCP startup failures, handshake failures, config errors, partial startup, and degraded mode need clean classification.
• ROADMAP.md:67 — Product principle §5: MCP partial success must be first-class and structurally report successful and failed servers.
• ROADMAP.md:1033-1059 — Phase 5: first-class plugin/MCP lifecycle contract and MCP end-to-end lifecycle parity.
• .omx/cc2/board.md Stream 5 active headings: CC2-RM-H0010, CC2-RM-H0080, CC2-RM-H0081, and CC2-RM-H0082 remain the goal-level source-of-truth anchors for plugin/MCP lifecycle maturity.
• PARITY.md harness checklist: mock parity scenarios are the executable regression surface for streamed model turns, plugin tool roundtrips, permissions, compaction metadata, and token/cost output.

Mock integration anchors

Area	Artifact/evidence
Deterministic model server	rust/crates/mock-anthropic-service/src/lib.rs implements the Anthropic-compatible mock server and scenario router used by CLI parity tests.
End-to-end CLI mock harness	rust/crates/rusty-claude-cli/tests/mock_parity_harness.rs starts the mock server, runs clean-environment claw commands, asserts JSON output, and optionally writes a machine-readable report via MOCK_PARITY_REPORT_PATH.
Scenario manifest / docs parity guard	rust/mock_parity_scenarios.json is required to stay ordered with harness cases; rust/scripts/run_mock_parity_diff.py --no-run verifies every manifest parity_refs[] string exists in PARITY.md.
Convenience runner	rust/scripts/run_mock_parity_harness.sh runs cargo test -p rusty-claude-cli --test mock_parity_harness -- --nocapture.
Plugin-path regression	plugin_tool_roundtrip loads an external plugin fixture from isolated settings and executes plugin_echo through the runtime tool registry.
Lifecycle-adjacent regression	auto_compact_triggered and token_cost_reporting prove runtime JSON keeps compaction and usage/cost fields parseable under mock responses, preventing parity drift in machine-readable output.
MCP degraded-startup regression	rust/crates/runtime/src/mcp_stdio.rs::manager_discovery_report_keeps_healthy_servers_when_one_server_fails proves a healthy MCP server remains callable while a broken peer is surfaced in a structured degraded report.
Plugin lifecycle state regression	rust/crates/runtime/src/plugin_lifecycle.rs unit tests cover healthy, degraded, failed, and shutdown states plus startup-event mapping.

Regression verification commands

Use the smallest command that proves the changed or audited surface, then broaden only when integration risk requires it.

• Mock scenario/docs map only:
  • cd rust && python3 scripts/run_mock_parity_diff.py --no-run
• Full mock integration:
  • cd rust && cargo test -p rusty-claude-cli --test mock_parity_harness -- --nocapture
  • cd rust && python3 scripts/run_mock_parity_diff.py
• Plugin/MCP lifecycle contract:
  • cd rust && cargo test -p runtime plugin_lifecycle -- --nocapture
  • cd rust && cargo test -p runtime mcp_stdio::tests::manager_discovery_report_keeps_healthy_servers_when_one_server_fails -- --exact --nocapture
• Standard Rust gates for implementation changes touching these surfaces:
  • cd rust && cargo fmt --all -- --check
  • cd rust && cargo check -p runtime -p rusty-claude-cli -p mock-anthropic-service
  • cd rust && cargo clippy -p runtime --all-targets -- -D warnings

Known gaps / follow-ups

• The mock parity harness validates plugin tool execution but does not yet spin up a real MCP stdio server through the CLI prompt path; MCP degraded-startup remains covered by runtime manager tests.
• Worker-4 owns the plugin command fallthrough regression implementation lane (task-10); this map records the verification/docs boundary and should not duplicate that parser work.
• Full cargo clippy -p runtime --all-targets -- -D warnings can be blocked by unrelated policy_engine.rs clippy violations in this worktree; when that happens, report the exact pre-existing diagnostics and keep focused lifecycle tests green.
• No .omx/ultragoal files were changed; leader-owned Ultragoal checkpointing remains outside worker scope.

Delegation evidence

Subagent spawn evidence: Task 9 spawned repository map probe 019e291d-e700-7171-b7bc-27ec0f6c850f, debug/root-cause probe 019e291d-e86f-78d0-a137-214ede03285c, and test/docs probe 019e291e-135c-79e1-80d0-9fd82866bd6e before deeper local inspection. The repository-map probe errored with 429; the remaining probes did not return before the local verification map was grounded from repo evidence, so direct findings above were integrated.