Read agent communication receipts without mistaking delivery for a human read
Find the current stage of a swarm message without confusing durable storage, delivery, runtime receipt and viewing. This guide connects local checks, the private observation screen and distinct receipts, so a delivery acknowledgement does not become an unsupported claim that an action completed.
🎯 For Tvijo operators and developers reviewing task events in an owner-scoped workspace or VS Code workflow.
- Separate persisted, queued, bridge-sent, runtime-received, and human-reviewed states.
- Use a cursor to resume a journal projection without treating a page view as approval.
- Recognize when runtime receipt is still unknown.
A journal can make task messages easier to replay, but receipt labels must stay precise. In the 9 Nov 2026 bounded implementation, permitted intermediate agent messages and tool start/result/error identities are appended to a durable journal and sent through the Worker’s existing outbox. The bridge’s ok/event_id response is a storage acknowledgement: it does not prove a runtime received the event or a person read it.
Goal
Inspect one task’s communication evidence and identify the strongest state supported by an explicit receipt. Keep the same task/account scope when moving between the VS Code task and an authenticated workspace view on a phone.
Offline contract command
From the repository root, run the deterministic contract tests:
python3 -m unittest tests.test_agentroy_communications tests.test_agentroy_worker
These tests exercise local persistence, cursor replay, coverage gaps, and a fake subprocess receiver. The fake receiver is not an installed DSHREM or Codex adapter. The read-only projection is intended for the existing owner-scoped Platform/RemDesk experience; UI wiring and live adapter receipts are not established by this test.
How to read the states
persisted: the journal transaction committed.queued: the existing outbox has delivery work.sent: the bridge/storage endpoint accepted the event.runtime_received:UNKNOWNuntil an explicit receipt contains the matching event ID.human_displayedandexplicitly_viewed_or_approved: remain unclaimed until the UI records those actions explicitly.- action completion: unknown unless a separate completion receipt says otherwise.
If you resume a feed, use its cursor and preserve the account/task filter. A repeated event ID with the same payload is idempotent; a changed payload under that ID is a conflict, not a new message. For editor context and connector concepts, see MCP in VS Code and what MCP is.
When this does not work
The test command should complete without a model/API call. A fake receiver can prove the receipt contract in isolation, not that a production runtime supports it. Common error: reading sent as “the agent saw it” or treating a successful GET as “the person read it.” Keep runtime and human states UNKNOWN until each has its own explicit evidence.
Claim scope: COMMS-01, implementation observed 2026-11-09. Local journal and fake receiver contract only; native production runtime receipt remains UNKNOWN; browser display is separately tested on synthetic data.
Inspect delivery in Platform
The /communications route was verified in an isolated candidate. Its production activation is currently BLOCKED: the running gateway loads from a shared checkout containing foreign unfinished work, and the current disk files have not been proven equivalent to the running process. We preserved that gateway instead of restarting it for this feature. The following actions describe the tested candidate with a synthetic signed-in account, not an already available public screen. Only explicitly assigned account journals are visible. An empty list means no assignment, not proof that no message was lost. Refresh reads the journal; next events use an ordered cursor. Visible cards can record a separate browser display receipt; the acknowledgement button records a distinct explicit client acknowledgement. Neither proves human attention or action completion. Working messages stay private.
python3 -m pytest -q tests/test_platform_communications.py
python3 scripts/agentroy_communications.py --help
Expected: eight local test cases without model API, plus CLI help. The owner configures a private registry and explicitly assigns journals; the tool refuses implicit reassignment of an existing ID. Do not infer ownership of another journal. Native Codex/DSHREM do not provide runtime receipt through this adapter: UNKNOWN remains visible. Isolated runtime receipt and real ASGI/SQLite/UI were checked on a synthetic account, not production working messages.
Phone and VS Code
From a phone, use Termius/SSH and Bash in a separate checkout; enter commands in the shell, not an agent task box. In VS Code Remote SSH use that checkout’s integrated terminal. Related setup: MCP in VS Code. Browser layout was checked at 320/390 pixels; a physical iPhone was not tested in this package.
Evidence
Checked 2026-11-09. Source files are available to maintainers in the private checkout containing this release; offline commands require that checkout, Node 26.7.0 and Python 3.12. Structured data and claim IDs · Primary-source audit · QM, reviewed commit · Promptfoo custom providers.
2026-11-09 UTC. The frozen pilot records and local scenarios listed above were checked. Until a separate publication receipt exists, a local build and screenshot do not establish that a public URL serves this release.
Practice
Run the audit-reading command above and preserve its output next to the check date. This reads published JSON without making a model request.
Check yourself
Inspect a synthetic receipt and locate its stable event ID, sender, recipient, run and cursor. Does storage ACK prove agent receipt? Does GET prove human attention? Both answers are no. With UNKNOWN, first establish adapter coverage and the explicit receipt instead of replaying an action. Compare state before and after reconnecting: event order must remain stable, and completion must not appear merely because the interface reloaded.
Next
Continue with the linked guides: evaluation, delivery and source verification. An expanded sample is planned only. New reusable skills remain hypotheses: evaluate their benefit on different tasks with a frozen method and independent acceptance.
Sources
Pilot data, ten-item audit, Promptfoo documentation, QM reviewed commit. Sources establish provenance; applicability and benefit depend on the conditions of the specific check.
COMMS-IMPL-01 — implementation scope. Checked 2026-11-09 UTC: the durable journal, outbox, cursor and separate client receipts work in the isolated implementation; production activation is not confirmed. Sources are independent local tests and saved synthetic ASGI/SQLite/browser receipts; working messages remain private. Journal source SHA-256 prefix 5f7c05159150. Contrary evidence and limits: native adapters do not provide explicit runtime ACK, and the live gateway base is not pinned. Recheck after a verified immutable gateway base and separate installation admission are available.
🔗 Article connections
Open the full knowledge graph ↗Published article links. Select a node to open its article at the heading.
Show the connection graph
Article links · keyboard / text view
Same topic
- Tvijo agent system: PRIZMA, RUNLEDGER and acceptance
- Audit an AI technology radar against primary sources before promoting a story
- Run an AI coding harness evaluation offline before any live pilot
- DSHREM on a phone: SSH, tmux, and safe task handoff
- DSHREM owner practice: cwd, worktrees, WIP, and root boundaries
- DSHREM economy swarm: checkpoints, routes, and real cost evidence
Next steps
Related reading
Tvijo agent system: PRIZMA, RUNLEDGER and acceptance
Tvijo is developing a controlled workflow from a repository task to an accepted result, with permissions, resource limits and attributable evidence. PRIZMA describes the intended resource profile; RUNLEDGER connects the work to its checks a…
AI and process foundationAudit an AI technology radar against primary sources before promoting a story
Check a technology news item against its primary source, version and event date before deciding to adopt it. This guide follows a fixed ten-item sample, separates facts from interpretation and forecasts, and preserves limitations and the condition for checking the claim again.
AI and process foundationRun an AI coding harness evaluation offline before any live pilot
Check the coding-agent evaluation path without a model request, then read eight real engineering pilot slots. This guide connects commands, independent artifact acceptance, full elapsed time and expense receipts, explaining why a saved response differs from a newly accepted result.
💬 Comments