Stacktrace · runtime detection
What Stacktrace detects, measures and blocks for Claude Code, Devin, Codex, the Cursor CLI, the Cursor IDE and Cursor cloud agents, and what cannot be supported for each agent and why. OpenAIDR 0.8.0 reads the Cursor CLI. OpenAIDR 0.9.0 reads the Cursor IDE's agent panel. Hooks that the installer writes read Cursor cloud agents.
Each cell is the short answer. The other tabs give the detail and the reason. Claude Code covers its CLI, IDE extensions and Claude Code on the web.
† waits on a field not yet confirmed in a real session. cost-personal-anomaly runs in Fleet on uploaded usage, so it works wherever the usage upload does. Codex grades are the built code's, verified end to end over synthetic Codex homes. Cursor CLI grades come from the stacktrace-demo policy suite, run through the real Cursor CLI and a stand-in model: 13 of 17 cases passed. Both rm -rf cases were refused by Cursor's Delete protection, which that build always turns on. Cursor IDE grades come from the same suite, run in the desktop app's agent panel: 15 of 17 cases passed on the finished sessions. blocked-quota and blocked-policy were missing, as on every kind. Cursor cloud grades come from agents run on Cursor-managed VMs on 2026-10-28, and from a simulated VM. A cloud grade with † rests on the simulated VM and the tests, not on a live run.
Claude Code's one gap is in a session on the web: a delete written with ~ or $HOME is not read, because /workspace is outside the home folder (ADR-0146). The Claude Code plugin and the playground are Claude-only by scope, not by a missing record.
Source of truth: docs/agent-support.md in stacktrace-ai/stacktrace. The Codex evidence is Codex CLI 0.163.0 run against a mock model and its source at rust-v0.163.0, and a live run on gpt-6.1-sol against a local Fleet. The Cursor evidence is the Cursor CLI 2026.10.19-6bbf2c4 run against a mock model and a stand-in model, and its shipped bundle, and the Cursor desktop app 3.27.12 run in its local inference mode against a stand-in model. The Cursor cloud evidence is agents run on Cursor-managed VMs from the API and from Cursor's dashboard, the stacktrace-demo live suite (12 of 12 checks), and its simulated VM suite against a local Fleet (26 of 26 checks). The Devin hook evidence is Devin's hook docs, the Devin CLI 3000.11.234 binary run with no sign-in, the plugin's hooks run against a real daemon by Devin's documented hook contract, and a signed-in Devin CLI 3000.11.234 session on 2026-10-31. Devin Desktop was not run. The block-finding grades rest on Codex's source at rust-v0.163.0 and the Cursor CLI 2026.10.22-27dcf5f bundle. The reader designs are docs/specs/codex.md and docs/specs/cursor.md in open-agent-security/openaidr.