feat(mcp): add stop_detached_probe to stop a listed probe by id

[jbachorik]

What & why

list_probes (re-enabled in #11) reports probes a client detached from, but the server couldn't act on them: exit_probe and detach_probe only reach probes the current server session deployed. This adds stop_detached_probe (pid, probe_id, optional port). It reconnects to the probe through BTrace's new public Client.connectAndExitProbe and stops it. probe_id accepts a bare id or a whole list_probes line (1: <id> [<class>]).

• The call is bounded by a 30 s timeout that closes the client, so an unresponsive agent can't hang the MCP request.
• Skills: btrace-mcp-operations and btrace-probe-lifecycle now hand a probe found by list_probes to stop_detached_probe, after the operator confirms it's no longer needed. The mcp-leftover-probe eval expects the new tool.

Building this exposed that reconnecting to a detached probe never worked in the BTrace agent (the CLI's -r <id> exit hung the same way). That fix and the client API are in btraceio/btrace#967.

How verified

Against a BTrace JAR built from the btraceio/btrace#967 branch:

Gate	Result
./gradlew check e2eTest (BTRACE_JAR=…)	unit 19 passed, 0 failed; e2e 11 passed, 0 skipped, 0 failed
scripts/validate-marketplace.sh	Marketplace validation passed (10 eval cases, 5 plugins, 31 skills)

Before the agent fix, the new e2e case hung for 90 s on the stop call; that's how the agent bug was found.

Coverage notes

• e2e stopDetachedProbeRemovesTheListedProbe: takes the detached probe's line from list_probes, stops it, requires its @Export counter to freeze, checks the probe drops off list_probes, and checks a second stop is an error.
• Unit: the schema and the missing-probe_id error; the reflective adapter call; parsing a list_probes line; and the timeout path (the client gets closed and an error is returned).

Not verified / known limitations

• Draft until fix(agent): keep detached probes reconnectable; add Client.connectAndExitProbe btrace#967 merges. The mcp-e2e workflow builds BTrace's default branch, which won't have Client.connectAndExitProbe before that; until then the e2e job is expected to fail with NoSuchMethodException.

Breaking changes / migration

None. This adds a tool.

Links

• Follows fix(mcp): re-enable list_probes now that the BTrace client no longer exits #11 (merged); rebased onto main.
• Depends on fix(agent): keep detached probes reconnectable; add Client.connectAndExitProbe btrace#967.

🤖 Generated with Claude Code

[jbachorik]

The failing e2e check is the expected dependency failure: the PR-triggered run builds BTrace develop (155b35d), which doesn't have Client.connectAndExitProbe yet (NoSuchMethodException). A manual run against btraceio/btrace#967 (btrace-ref=feat/client_stop_probe, bdcf0ef) passes: https://github.com/btraceio/agent-plugins/actions/runs/37290693610. Re-run once #967 merges.