test(core): detach handler, evaluate default-frame, adoption edges + threshold ratchet

[debugmcpdev]

Closes out the coverage-expansion sprint (after the source-map-derive rewrite and the adapter pack). Three cohesive core gaps plus the threshold ratchet.

server.ts: the detach_from_process handler had zero tests

The entire handler body (~40 lines, the largest single uncovered block in server.ts) was untested. Added dispatch tests: success payload shape, success:false propagation, terminateProcess defaulting, the session-state McpError catch mapping to {success:false, state:'stopped'}, rethrow of unrelated errors, and the missing-sessionId guard.

session-manager: evaluateExpression default-frame resolution

The dense uncovered cluster was the frameId-omitted path. New suite pins: the not-paused guard, missing current-thread failure, top-frame anchoring via a stackTrace request (asserting the evaluate call carries the derived frameId), empty-stack failure, and stack-trace-rejection wrapping.

child-session-manager: adoption edges

• The child-safe policy (previously 0%): children receive a policy with reverse-start disabled and, crucially, a handleReverseRequest wrapper that acknowledges-and-stops grandchild startDebugging requests while passing unhandled requests through — asserted via the policy the mocked child captures.
• Adoption catch arms: failing setExceptionBreakpoints/setBreakpoints/configurationDone don't abort adoption.
• handlePostAttachInit: a post-attach initialized replays stored breakpoints (new mock knob, mirroring emitStoppedAfterAttach).
• The js-debug threadId === 0 double-pause quirk; threads-failure tolerance; death-latch re-entrancy (error followed by close); shutdown surviving a child whose own shutdown() throws.

Threshold ratchet

vitest.config.ts coverage thresholds move from statements: 80 (10+ points of dead slack — nothing was regression-guarded) to statements: 90, branches: 80, against measured post-sprint levels of 92.9 / 83.0 (margins absorb platform-specific branches).

Sprint result (full coverage run on this branch)

	before sprint	after
overall lines	90.6%	93.3%
statements	90.0%	92.9%
branches	80.4%	83.0%
functions	88.3%	91.0%

Every sprint target is off the deficit leaderboard: source-map-derive (was #1 at 8%) now 98.5%; the mock adapter (65%) and the go/rust/python adapters and child-session-manager no longer chart. The remaining top entries are the deliberately out-of-scope scatter (server.ts 91%, session-manager-operations 93%, dap-proxy-worker 91%).

Full npm test green, npm run lint clean.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!