Files
prose-contracts/zulip-mention-reliability.prose.md
T
tanko-bot c23462eba8
PR Pipeline — Authorize → Validate → Review → Merge / auth (pull_request) Successful in 4s
PR Pipeline — Authorize → Validate → Review → Merge / validate (pull_request) Successful in 1s
PR Pipeline — Authorize → Validate → Review → Merge / lint (pull_request) Successful in 1s
PR Pipeline — Authorize → Validate → Review → Merge / ai-review (pull_request) Successful in 1s
PR Pipeline — Authorize → Validate → Review → Merge / gate (pull_request) Successful in 1s
fix: correct tanko runtime to DSH across contracts & scripts
Tanko migrated from Hermes to DSH (DeepSeek Harness) on 2026-08-27. Update all
records that described tanko as a Hermes agent / Hermes runtime:

- infra-control: CT 112 tanko platform Hermes -> DSH
- zulip-health / zulip-self-heal / zulip-mention-reliability / pi-approval:
  tanko is on DSH, mumuni remains on Hermes
- memory-audit-maintenance: exclude tanko from Hermes roster (uses DSH-native memory)
- hermes-config-template / hermes-agent-baseline: remove tanko from Hermes roster,
  keep LiteLLM key alias 'tanko'
- hermes-zulip-plugin / hermes-zulip-restore / build-zulip-plugin: tanko excluded
- infrastructure-maintenance: gateways check no longer probes Hermes on tanko CT112
- scripts/daily-infra-report.py: fix CT-ID regression (CT 122->112), report tanko as DSH
- scripts/zulip-monitor.sh: stop probing tanko's retired Hermes gateway
- scripts/agent-health-check.py: skip Hermes gateway checks for tanko (runtime=dsh)
- scripts/prose-auth-check.sh + AGENTS.md: authorize tanko/tanko-bot for its own records

Tanko remains CT 112 at 192.168.68.122; infrastructure facts unchanged.
Dated/incident records (run logs, migration logs) left intact as history.
2026-08-27 03:01:24 +00:00

3.1 KiB

kind, name, status, description
kind name status description
responsibility zulip-mention-reliability retired RETIRED 2026-07-04 — Zulip extension decommissioned. Abiba now operates exclusively through Telegram (@AbibaBot). This contract is kept for historical reference only.

⚠️ RETIRED — The pi Zulip extension (~/.pi/agent/extensions/zulip/) and PM2 process (abiba-zulip) have been decommissioned. All mention/reliability monitoring now happens through Telegram. Agents (Mumuni on Hermes, Tanko on DSH) and Agent Zero (kagentz) continue to use Zulip.

Maintains

  • mention_response_rate: number — % of @mentions that get a response (target: >95%)

  • last_test_result: "pass" | "fail" | "degraded"

  • known_failure_modes: array — History of what went wrong and how it was fixed

  • queue_health: "healthy" | "expired" | "reconnecting"

    Postconditions

    • test_mention_gets_response == true — Send @Abiba Bot test, confirm response within 30s
    • stream_subscription_active == true — Bot is subscribed to #agent-hub and #general chat
    • queue_active == true — PM2's event queue is registered and polling
    • no_echo_loop == true — Bot doesn't respond to its own messages
    • dm_always_works == true — DMs still respond even if streams are broken

Known Failure Modes (from History)

Failure Root Cause Fix
No stream events Bot had 0 stream subscriptions Subscribe via API
No @mention detection botUserId was null (register endpoint doesn't return it) Fetch from /api/v1/users/me
No stream events zulip-js library sends multipart/form-data instead of form-urlencoded Use URLSearchParams directly
Double responses Both TUI and PM2 loaded the extension ZULIP_EXTENSION_ACTIVE guard

Remediation Rules

Rule 1: Queue Expired

  • Detect: Poll returns BAD_EVENT_QUEUE_ID
  • Fix: pm2 restart abiba-zulip
  • Verify: Check logs for "Connected, queue:"

Rule 2: Subscription Missing

  • Detect: GET /api/v1/users/me/subscriptions returns 0 for #agent-hub
  • Fix: Subscribe via POST /api/v1/users/me/subscriptions
  • Verify: Re-check subscriptions

Rule 3: Extension Crashed

  • Detect: PM2 status shows "errored" for abiba-zulip
  • Fix: pm2 restart abiba-zulip
  • Escalate: If restarts >5 in last hour, alert user

Rule 4: No Response to Test Mention

  • Detect: Send test @mention, wait 30s, check if response appeared
  • Fix: Run diagnostics (check queue, check code, check logs)
  • Log: Record failure mode as new entry in Known Failure Modes

Continuity

  • On user report: User says "you didn't respond" → run diagnostic
  • On test failure: Test mention fails → run remediation
  • Retrospective: Every 6 hours — send a test @mention to self-verify
  • On PM2 restart: Always run a self-test after restart

Execution

  1. Self-test: Send @**Abiba Bot** _selftest_ in #agent-hub, wait for response
  2. If no response in 30s: Run diagnostics
  3. Diagnose: Check queue, subscriptions, PM2 status, logs
  4. Fix: Apply matching remediation rule
  5. Retest: Try again
  6. Log: Record success or new failure mode
  7. Report: DM user only if escalation needed