v2.1.25

Commits

• 7eaa834 release: 2.1.25
• 2b57a17 test(F-950): the real-Chrome re-attach pin writes the record path as the registry does, so it names the master on macOS
• 3e35a85 test(F-950): the driven-snapshot pin spells the master path as the record does, so it names the same directory on macOS
• 3c9d4ea fix(F-950): bisect only Chrome's refusal, never loop on an empty jar, word the claim refusal for us
• 92ec136 fix(F-950): one cookie Chrome refuses no longer drops the whole hand-off jar
• 44b2318 fix(F-950): a browser this backend re-attached to is not read as a sibling backend's
• d2644f0 test(F-949): the slow-init stand-in takes the keyword the proxy now passes, so the thread it pins is alive and not dead of a TypeError
• ac07768 docs(F-946): one blank line between the Unreleased entries
• ee0fb95 docs(F-946): state the two tools/call wire changes; name where fastmcp 3's argument DEBUG line moved
• 3558522 docs(F-946): CHANGELOG entry under a fresh Unreleased, above 2.1.24
• 7b95444 test(F-946): pin that a docstring is served once, as the whole description
• c6e5fc6 fix(F-946): keep Args: text out of the tool schemas; correct the lifespan claim
• d7b2907 fix(F-946): move to fastmcp 3.x, clearing the last fastmcp advisories
• 1c791f9 fix(F-949): the proxy's Sentry init no longer auto-enables integrations that import mcp while the main thread does
• da78f62 test(F-947): the future-stamp heartbeat pin stamps a whole bound past the edge