Agent strategy inbox contents — what shows up and how to act on it
The Aura sentinel inbox surfaces strategy messages between agents — task handoffs, conflict alerts, plan amendments. Full schema and how to triage from CLI or MCP.
When you run more than one AI coding agent against the same repo (Claude in one terminal, Gemini in another, a Sentinel watcher in the background) they need a shared mailbox. Aura's agent strategy inbox is that mailbox.
What lives in the inbox
- Strategy messages. "I'm about to refactor
auth/middleware.rs; pause on that file for 10 minutes." - Plan amendments: "Wave 4 was unnecessary, skipping; reordering wave 5 to land before wave 3."
- Conflict alerts: "Claude touched
handlers/billing.rsat line 142; Gemini's pending edit at line 138 will collide." - Handover packets: dense XML payloads from
aura_handoverwhen one agent compacts state for another to resume. - Sentinel decisions: release approvals, gate verdicts, autonomy-budget adjustments.
Schema
{
"id": "01HXY…", // ulid
"from": "claude:wave-3",
"to": "gemini:wave-3" | "broadcast" | "sentinel",
"kind": "strategy" | "plan" | "conflict" | "handover" | "decision",
"subject": "pause on auth/middleware.rs",
"body": "…", // free text or structured JSON
"refs": { "files": [...], "blocks": [...], "tasks": [...] },
"created_at": "2026-04-26T14:02:11Z",
"signed_block_id": "01HXY…" // every message lands as a signed block
}Reading the inbox
# Tail unread aura sentinel inbox --unread # Filter to strategy messages from a specific agent aura sentinel inbox --kind strategy --from claude # From an MCP-aware client (Claude Code, Cursor) aura_sentinel_inbox(unread=true)
Why every message is signed
A strategy message that says "skip the security review on wave 7" is dangerous if it can be forged. Every inbox entry is wrapped in the same signed-block envelope as Aura's intent log, so the receiver can verify the sender's key chain before acting on the instruction. Tampered messages fail aura attest verify.
What is in an agent strategy inbox?
Messages between agent sessions that change what another agent should do: task handoffs, collision alerts, and amendments to a plan already in progress. It is distinct from a log — every entry is meant to be acted on or dismissed.
How should an agent triage its inbox?
Collisions first, because they get more expensive the longer you edit. Then plan amendments, which may make the current task obsolete. Handoffs last. Anything read and not acted on should be closed explicitly so the next session does not re-read it.
Aura Crew vs Claude Code Loops: the autonomous work-loop, compared
Claude Code defines four loop types — turn-based, goal, time, and proactive. Aura Crew is the same idea productized: a dependency-ordered work-loop with proof, collision-safety, and any agent. Full comparison.
Autonomous coding agent loops, explained: turn-based, goal, time, and proactive
A coding agent loop is an agent repeating cycles of work until a stop condition is met. The four types explained — turn-based, goal-based, time-based, proactive — with when to use each and how to keep quality high.
How to run multiple coding agents in parallel without merge conflicts
Running several AI coding agents at once collides on shared files. The fix: a worktree per agent, a live team radar, and soft-to-hard zone claims so parallel agents never touch the same symbol. How Aura Crew does it.
Goal-based agent loops: giving a coding agent a definition of done
A goal-based loop keeps a coding agent iterating until a success criterion is met. The trick is a deterministic definition of done — tests, a score, a proof — not an LLM guessing at "good enough." How to write one.