--- name: owl-send-binary description: "Inter-agent messaging: use spt-core `spt send ` (SENT/QUEUED = success); $OWL send is the LEGACY plugin owl.exe queuing into the legacy home — a black hole for spt-core agents (doyle-diagnosed 2026-07-10)" metadata: node_type: memory type: reference originSessionId: 975b3c8a-3321-487c-ae8b-b9df8cf13258 --- **REVERSED 2026-07-10 (doyle-ruled):** reach agents via `printf '%s' "" | spt send ` — the spt-core binary (`$LOCALAPPDATA/spt-core/bin/spt`). `SENT` = live delivery, `QUEUED` = spooled for their next drain; both are success, never retry. Doyle: "ALL agent-bound traffic goes `spt send` from now on, retire the $OWL habit." **`$OWL send` is LEGACY** (plugin `owl.exe` v1.11.x) — it queues into the legacy/modern-spt home (`$LOCALAPPDATA/spt`), which spt-core agents NEVER drain. On 2026-07-10 every `$OWL send doyle` black-holed there ("CAPACITY_WARN: 1193 queued" = that dead queue) while doyle's spt-core spool had zero rows from me; `spt send doyle` returned SENT and landed instantly. The old NO_PERCH-from-`spt send` advice below the fold dated from BEFORE the agents migrated to spt-core endpoints (~2026-07-06); it no longer applies. Two parallel messaging systems share this box during the parity build-out (homes `spt-core/` vs `spt/`). The live-agent roster that matters (doyle, perri, flynn, lia, deployah, todlando…) is spt-core's. **Fallback channel to doyle** when messaging breaks: write `.claude/todlando-to-doyle.md` at the repo root — he polls it. See [[cross-machine-test-rig]], [[v013-rc-attach-flood-wedge]] (doyle = forkpty CI partner).