Why
With several agents across several machines, the question is never "what is running", it is "which one is waiting on me". CC Pocket aggregates pending approvals (commands, file edits, MCP requests, agent questions) across sessions and it is the best thing in that product. Nobody in the field does it across machines. flue already has the fleet, so this is home turf.
Rough scope
- Detect "waiting on input" per session: a Claude Code permission prompt, a Codex approval, a plan-mode question, or a quiet terminal sitting at a prompt after a long run.
- Detection can start from what flue already has: the agent transcripts parsed for the agents screen, plus terminal quiescence heuristics. It does not need to be perfect on day one, it needs to be useful.
- Surface: a badge count in the shell of the app, a filter or section at the top of the sessions list, and the same state visible in the ⌘K switcher. One tap lands in the session at the prompt.
- Feeds the push notifications issue: "session needs you" is the highest value push there is.
Notes
- Claude Code and Codex both render prompts in the terminal, so the terminal stream is the source of truth even when transcripts lag.
- False positives are cheap (you glance and move on), false negatives are expensive (agent sits blocked for an hour). Tune accordingly.
Why
With several agents across several machines, the question is never "what is running", it is "which one is waiting on me". CC Pocket aggregates pending approvals (commands, file edits, MCP requests, agent questions) across sessions and it is the best thing in that product. Nobody in the field does it across machines. flue already has the fleet, so this is home turf.
Rough scope
Notes