v1.2.0

Everyone gets these changes on update; nothing here is opt-in.

Fixed

• Agento finds the Claude Code CLI where it is actually installed. A GUI launch does not inherit your shell's PATH, so a working claude could still be reported as missing — and agents genuinely could not run. Detection now probes your login shell, honours a claude migrate-installer install in ~/.claude/local (which is a shell alias, not a binary on any PATH), and checks the known install locations. You can also set the path explicitly in Settings. (#503)
• An integration no longer hosts zero tools in silence. A service saved without its own tool list exposed nothing to agents while still advertising the tools, and unticking a service's last tool in the UI meant "expose everything" — the exact inverse of what the screen says. Both are fixed, and a run now logs which tools each MCP server actually registered. (#501)
• A resumed Claude session shows its history, and is no longer continued twice. "Continue in chat" is idempotent per source session. (#490, #492)
• A mcps.yaml in your data directory no longer disables every MCP-backed agent. Its mere presence used to refuse those turns; the file is now read directly, with the three transports and ${ENV:…} substitution. (#375)
• A cron task's expression is validated when you save it, so an expression that cannot parse or would never fire is refused at the form instead of failing silently at fire time. (#330)
• Windows: the directory picker, the Claude settings surface and the config-dir probe answer properly instead of returning 501. (#374)

Added

• Session journey. A session's timeline segmented into turns, with each sub-agent's steps nested under the Task that spawned it. (#479)
• Advanced session filters the backend already served are now in the UI. (#495)

Installing

Builds are neither Apple notarised nor Windows code signed, so the first launch of a downloaded build needs a Gatekeeper or SmartScreen bypass — see Installation. In-app updates are unaffected.