[FEATURE]: Agentic updates

[cybertruck021]

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

Subagents sometimes see the scan status and start doing extra work that isn't part of their responsibilities. I'd like to see agent updates similar to those in Claude Code, so that agents can communicate with each other and focus only on their own tasks.

[s0ld13rr]

Thanks for the detailed report - this is exactly the failure mode we targeted in 0.2.4, so you were almost certainly hitting it on 0.2.3.

Root cause you hit: on 0.2.3 every subagent got the entire multi-host engagement state (the full scan status) re-injected into its prompt each turn. Seeing everything, they'd wander outside their assigned task and redo work that wasn't theirs.

What 0.2.4 changes:

1. Full state is now gated to the coordinator only. A focused subagent no longer receives the multi-host scan status — it gets only its task target plus a one-line pointer to pull specifics on demand via state_query. This directly removes the "subagent sees the whole scan and starts extra work" trigger.
2. Prompts thinned to specialty-only. Each subagent's prompt was cut down to just its own responsibility (the shared cross-cutting rules moved into one kernel), so there's far less pull toward off-task behavior.
3. Plan-and-Execute doctrine. Specialists run a tight try → observe → reflect loop in a clean, short-lived context and return a condensed ; the coordinator integrates it and re-plans. Specialists don't inherit each other's transcripts.

On the "agent updates like Claude Code" part — that channel already exists and got reinforced in 0.2.4:

• Subagents surface critical findings to the coordinator via interrupt alerts (priority: "interrupt"), which a watcher drains into the coordinator between turns.
• Completed subagents return a condensed summary that the coordinator folds back in — instead of everyone reading the shared raw state.

Could you retest on 0.2.4 and let me know if the off-task behavior still shows up? If it does, the specific subagent + a snippet of what it wandered into would help us tighten the gating further. Keeping this open for that feedback.

[s0ld13rr]

@openredteam could you please re-check the issue with latest version?