Published
AgenticBench
Independent, repeatable tests of what AI coding agents actually do on your machine. The scorecard is updated as new agents, versions and tests are added: every result links to its evidence, vendors hear about findings before publication, and corrections are dated. The test rig is public (github.com/agentic-thinking/agenticbench, Apache 2.0). The efficiency series covers the same agents; vendor-hosted agents are shown separately because they could not run on the fixed model.
What agent harnesses record, and what they send home
The dated write-up behind AgenticBench: what each harness records on your machine, what it sends off it, and how we test both.
Two labs, one failure: what 2026's agent incidents say about discovery
OpenAI and Anthropic had technically different containment failures in their 2026 agent evaluations. Comparing the same five dates across the year's disclosed incidents shows the common failure: operators learned of their own agents' activity weeks to months after it began, although the evidence largely existed.
The 2026 agent incidents: who held the record?
A sourced timeline of the OpenAI evaluation-agent breakout, the Hugging Face intrusion, the Australian Medicare statistics portal breach and related events, plus five shorter notes. Finding: every party held a fragment and nobody held a joined-up record, and the gap between an incident and it being reported publicly ran from 3 days to about 4 months.
What 11 agent runtimes let you record
A snapshot, dated 25 September 2026, of the documented hook surface of 11 agent runtimes, including Claude Code, Codex CLI, Gemini CLI, Cursor, GitHub Copilot CLI and the OpenAI Agents SDK, scored against the AgentHook v0.2 tiers. Finding: only Hermes documents enough to be Gold-capable, and only when reasoning capture is enabled; five, including Claude Code and Codex CLI, expose no hook around model calls at all.
Evidence lost in transit
We replayed synthetic incident scenarios through two HookBus builds. On the earlier release, the governance fields needed for reconstruction were dropped in transit (0 of 75 delivered), and a secret exfiltration split across three individually permitted steps went undetected. With the fix, 75 of 75 arrived and the sequence was flagged. A small, synthetic test: it frames the question rather than answering it.
AgentHook: A Runtime Evidence Standard for Auditable AI Agent Governance
Ruocco, P. (Leo). Draft v0.2, May 2026. Defines a vendor-neutral record of agent lifecycle events, actions, decisions and approvals, with conformance tiers.