wazionapps/nexo

v7.17.4 Bugfix

This release fixes issues for SREs watching stability and regressions.

Published 23d MCP Developer Tools

View tool

✓ No known CVEs patched

Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

ai-memory atkinson-shiffrin autonomous-agent claude-code claude-code-plugin codex

+14 more

cognitive-architecture knowledge-graph local-first mcp mcp-server memory metacognition model-context-protocol natural-forgetting llm semantic-search shared-brain trust-scoring vector-db

ReleasePort's take

Light signal

editorial:auto 13d

Automation runners now properly differentiate full NEXO agents from strict JSON children, eliminating protocol contamination in outputs. Additional telemetry and doctor improvements enhance observability.

Why it matters: Patch to v7.17.4 immediately if your environment uses automation runners; the fix prevents corrupted JSON payloads that could break downstream processing.

Summary

AI summary

Automation runners now correctly distinguish full NEXO agents from strict JSON children, preventing protocol contamination.

Changes in this release

Type	Severity	Summary	CVE
Feature
Feature	Medium	Automation telemetry records caller contract metadata for support visibility. Automation telemetry records caller contract metadata for support visibility. Source: llm_adapter@2026-05-21 Confidence: high	—
Feature	Medium	Runtime doctor compares crons against automation runs, surfacing coverage drift. Runtime doctor compares crons against automation runs, surfacing coverage drift. Source: llm_adapter@2026-05-21 Confidence: low	—
Feature	Medium	Runtime doctor surfaces cron/caller coverage drift as a support warning. Runtime doctor surfaces cron/caller coverage drift as a support warning. Source: granite4.1:30b@2026-05-24-audit Confidence: low	—
Bugfix
Bugfix	Medium	Strict automation children no longer receive global protocol prompt contaminating JSON output. Strict automation children no longer receive global protocol prompt contaminating JSON output. Source: llm_adapter@2026-05-21 Confidence: high	—
Bugfix	Medium	Guardian metrics now correctly count injection events preventing zero capture rates. Guardian metrics now correctly count injection events preventing zero capture rates. Source: llm_adapter@2026-05-21 Confidence: low	—
Bugfix	Medium	Automation runners distinguish full NEXO agents from strict technical children. Automation runners distinguish full NEXO agents from strict technical children. Source: granite4.1:30b@2026-05-24-audit Confidence: low	—
Bugfix	Medium	Guardian metrics now count the real `injection` event alongside `enqueue` and `inject`. Guardian metrics now count the real `injection` event alongside `enqueue` and `inject`. Source: granite4.1:30b@2026-05-24-audit Confidence: low	—

Full changelog

[7.17.4] - 2026-05-11

Fixed — automation discipline contracts and Guardian observability

Automation runners now distinguish full NEXO agents from strict technical children. Full background jobs such as email_monitor, followup_runner, sleep/nightly, evolution/run, daily_self_audit, immune/scan, postmortem_consolidator and catchup/morning keep the complete task/evidence/diary/learning/followup discipline, while strict JSON children such as deep-sleep/extract, deep-sleep/synthesize, morning_agent, learning_validator and check_context no longer receive the global protocol prompt that can contaminate JSON output or create post-close loops.
Claude and Codex automation telemetry now records the caller contract. automation_runs rows carry caller/session/contract metadata consistently, including the Codex branch, so support can see whether a real background job ran as a disciplined agent or as a strict child owned by a parent job.
Guardian metrics now count the real injection event. The daily aggregator accepts enqueue, inject and injection, preventing capture-rate metrics from reading zero while Desktop/Brain telemetry is actually firing.
Runtime doctor now surfaces cron/caller coverage drift. Core automation crons are compared with matching automation_runs callers; missing caller-attributed rows become a support warning instead of a silent blind spot.
Coverage: targeted runner/metrics/doctor checks pass (175/175), lifecycle/email/productization sweep passes (193 passed, 2 xpassed), evolution/skills/watchdog passes (60/60), and live runtime evidence shows caller-attributed morning_agent dry-run plus Guardian metrics capture_rate=1.0.

Full Changelog: https://github.com/wazionapps/nexo/compare/v7.17.3...v7.17.4

View diff on GitHub

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Share on X Share on Bluesky

Track wazionapps/nexo

Get notified when new releases ship.

About wazionapps/nexo

Cognitive memory for AI agents with Atkinson-Shiffrin memory model (STM/LTM/sensory register), semantic RAG, Ebbinghaus decay, trust scoring, and 76+ MCP tools.

All releases →