wrapper 2026-10-10 wrappermcpclaude-codejev

rea-jev: System-1 reflexes around REA in Claude Code

REA already speaks MCP. rea-jev adds a **cheap reflex layer**: typed Jev decisions at the moments where a wrong agent impulse is expensive — before a risky tool call and before declaring done.

Source meta

0 GitHub stars (snapshot)

Language: JavaScript

License: MIT

Updated: 2026-10-07

GitHub source ↗

rea.run rating

4/5 — Best third-party REA productization

Quality. Clear System-1 hooks around REA MCP; fail-open design.

Evidence. Hooks route/gate/score/stop; evidence ledger; jev doctor.

Limits. Young plugin; thresholds may be uncalibrated; needs TypeSafe Jev.

For. Claude Code users who want reflexes around risky REA calls.

What it is. A MIT Claude Code plugin bundling the reverse-engineer skill, REA MCP (version-pinned), five hooks, a jev CLI, and investigator/verifier subagents.

Relation to REA. Claude stays System 2; Jev answers tiny routing/gating/evidence/stop questions in ~300 ms. Every hook fails open so a Jev outage does not brick analysis.

Workflow. /rea-jev:setup → investigate with MCP tools → PostToolUse scores evidence into a ledger → Stop hook blocks unverified “done.”

Evaluation. Clearest third-party packaging of REA we reviewed. Star count is near zero today; treat as early but architecturally thoughtful. Calibrate thresholds before trusting stop-gates in production labs.

Compliance

Plugin for authorized local analysis; does not change REA’s legal boundary.

Takeaways

Fail-open hooks preserve analysis when reflexes err.
Evidence ledgers make “done” falsifiable.
Pin REA + Jev versions before team rollout.

Related on rea.run