News
Incident reports, release notes, and project updates.
Agenthood v3.75.0 makes the OpenCode Zen provider protocol-aware and adds support for Jev System One.
Agenthood Team
Agenthood v3.74.0 adds an OpenCode client session backend for agenthood run, expanding LLM session options.
VS Code extension reaches v1.0.0 for Marketplace publish; patch v3.72.1 ships with an empty changelog.
v3.71.1 fixes an Opencode bug where bare plugin specifiers could crash the host process.
Agenthood v3.71.0 adds a goal command, origin-based concurrency arbitration, and one-step trajectory persistence.
Covers v3.69.0–v3.70.2: confidence patterns, context budgeting, and release pipeline hardening.
Herald release-note fixes, safer Dependabot grouping, and a Vitest 5 upgrade across three patch releases.
v3.68.0 ships Convert & Trust: onboarding, fail-closed skill trust, cost observability, plus dependency fixes.
A roundup of September 2026 AI developments — GPT-6 Astra, Claude Opus 5.5, Gemini 3.8, OpenAI's IPO push, EU AI Act enforcement era, and what to watch in October.
Agenthood v3.67.0 adds provider health probes, request timeouts, memory TTL, and tool safety; v3.66.0 improves CLI, init, and skill workflows.
Agenthood ships cross-provider eval comparison, packaged skill catalog entries, and a round of LLM, CI, and security fixes.
Bug fixes for verify placeholder scanning and marked options compatibility.
Agenthood v3.63.0 adds cost, doctor, rotate-key, ritual, and skill-integrity commands; v3.63.1 cleans up release tooling.
Evals gains description optimization, blind A/B, regression testing, and benchmark output; runtime checkpointing and resume improve.
New skill registry, checkpointing, and CLI tools, plus security and reliability fixes across v3.49.0–v3.56.6.
A roundup of Agenthood releases from September 4–5, 2026, covering skills CLI, MCP server, model defaults, CI reviews, plugins, and logging.
New ask_human park primitive for park-and-resume hosts, plus a test stub provider.
Patch release hardening LLM provider failover priority, per-entry keys, and retry behavior.
Three releases: agent count fix, LLM logging cleanup, runtime telemetry, and audit hardening.
v3.38.0 reframes steward load routing, adds the-mediator, and fixes a CI checkout issue.
Three releases tighten delegation, add conversational styles, and complete mind virus defense.
v3.35.1 stops milestones from editing CHANGELOG.md and release notes, letting semantic-release own those files.
Agenthood v3.35.0, v3.34.1, v3.34.0, and v3.33.1 ship a broader skills library, trust-boundary hardening, and observability fixes.
From redacted trace envelopes to anomaly alerts and Sentry reporting, Agenthood can now explain itself — and the Society welcomes ten new utility skills.
Release digest for v3.28.0–v3.33.0: observability, anomaly detection, eval improvements, and CLI updates.
Agenthood adds eval suites, trace persistence, cost estimators, and a hardened CLI in the latest release train.
Agenthood ships provenance store, precedent search, graph snapshots, and hardened CI/herald in three releases.
Seven releases from Aug 5–8: a new builder member, CI and review hardening, LLM rate-limit fixes, and RAG caching.
After PR #360, The Warden and The Auditor caught 3 issues humans missed — type duplication, silent error paths, and brittle tests — and filed 2 follow-up issues.
Agents can now use tools during chat — web_fetch for retrieving content and code_execution for running JavaScript in a sandboxed environment.
Streamlined the CI pipeline by removing a duplicate workflow, skipping builds on doc-only changes, and fixing stale-comment encoding bugs in reviewer automation.
A comprehensive audit of 106 documentation files found and fixed 43 issues — broken links, stale references, outdated paths, and inconsistencies across 21 files.
Stripped the init command down to essentials, fixed 5 quality findings from automated code review, and registered a new Society member — The Inspector.
The Oracle — a floating AI companion with quick actions for summaries, knowledge checks, and feedback — now on every docs and academy page.
A roundup of June 2026 AI developments — GPT-5.6, Siri AI, Claude Mythos 5, Grok 4.3, Gartner's ---.59T spending forecast, policy shifts, and what to watch in July.
New features: feedback API for chat responses, news section with RSS feed, enriched analytics, structured logging across the stack, and hardening follow-ups.
We've completed a security audit of the Studio playground — adding SSRF protection, CSP headers, origin validation, rate limiting, hydration fixes, and eliminating all lint warnings.
We've added Vercel Speed Insights to monitor real-user Core Web Vitals — LCP, CLS, INP, and TTFB — across all pages.
A detailed post-mortem of the Academy outage on April 28, 2026, covering the docs rendering issue, DNS misconfiguration, fixes applied, and preventive measures.