Relay-garble regression: @-mention corruption on outgoing relays (2026-06-16)

[OneStepAt4time]

Summary

Relay garble pattern from 2026-06-08 has regressed. Outgoing relays are corrupting @-mentions again (observed: @AAAAAAAAAAAAthena, @Hermess on a Boss directive dispatched to #aegis-devs at 2026-06-16 06:07 Rome).

Evidence

• Boss's 06:07 Rome directive to #aegis-devs was relayed with garbled @-mentions in the receiving agents' views
• Cross-actor pattern (multiple agents see the same garble simultaneously)
• Substantive directive content is correct (PRs, PIDs, ports all verify) — only the @-mention layer is corrupted
• The receiving agent (Hephaestus) correctly held execution and pinged PM for verification per the 2-safety-net rule from the original incident

Relationship to #4615

#4615 (filed 2026-06-08 20:21 by Hermes) documented the same pattern. The patch did not hold — this is a regression 8 days later. Linking for cross-reference.

Impact

• Substantive directives can still be executed correctly because the content layer is intact
• @-mention routing breaks, so:
  • Agents reading the relay may not see their names
  • Discord notifications don't fire on garbled @-mentions
  • Coordination depends on agents recognizing the directive by content, not by tag

Lane

Out of any agent's lane (shared OpenClaw infrastructure). Themis correctly stayed out of the original incident per Boss directive. PM filing this as a follow-up to #4615 because relay-garble triage is the PM's lane (the original triage comment at 20:28 was a PM comment).

Acceptance criteria (for whoever fixes the relay)

• Outgoing @-mentions arrive intact on all visible agents
• No @aaaaaaaaaaaa[username] or @Hermess-style corruption on subsequent directives
• 7-day no-regression window before closing
• No IC action required from the team

[OneStepAt4time]

Triage pass complete (2026-06-16, ~07:06 Rome).

Labels set: P2, needs-human (removed ready, needs-triage).
Board: added to Project #2, Stage = Backlog.

Lane: PM-tracked, OpenClaw infrastructure (out of any IC lane — per body, "No IC action required from the team").

Why P2, not P1: substantive directives still execute correctly (content layer intact); only @-mention routing breaks. Coordination friction, not blocker.

Acceptance window: 7-day no-regression hold before close (per body). PM will check daily at heartbeats.

Cross-references:

• infra(relay): bot account-mapping drift on relay messages (Hermes + Scribe confirmed) #4615 (2026-06-08 original, 8 days ago) — patch did not hold
• Working assumption: upstream OpenClaw fix; will surface on #aegis-devs only if a Boss-directed action becomes available

[OneStepAt4time]

Filing a sibling aegis-side tracker for the Layer-2 (sender-metadata envelope) and Layer-3 (reviewer-state) aspects of the relay-drift family — these weren't covered by the umbrella @-mention body and aren't in any agent's lane for the underlying OpenClaw fix.

→ #4828 — aegis-side defensive normalization for Layer-2 (sender-metadata) + Layer-3 (reviewer-state) relay-garble (P3, Hep lane).

#4828 is the aegis-side defensive layer (small slice: tests + additive normalization, mirrors #4615/#4617 + F-6 redacted pattern). Sibling to this OpenClaw-infra tracker.

Per Argus 2026-07-04 12:51 Rome (msg 1522917889538003055) review-lane nudge: filed because the current defense is manual gh pr view --json verification — undocumented invariant, future regression with no detector.