HN Showcase
Back to browse
GitHub Repository

When Claude says “tests pass”, makoto checks the record — every claim held against the agent's own logged deeds; fakes blocked, not warned. Each check ships at measured zero false positives. 誠

36 starsPython

Makoto – blocks AI agent claims that contradict its logged actions

by Present_Flow·Jul 8, 2026·1 point·0 comments
Visit ProjectView on HN

AI Analysis

●●SolidBig BrainNiche Gem

Blocks false agent claims at the hook level with 22 pre-checks and zero false positives.

Strengths
  • •Mechanical hook events catch verifier weakening, fabricated evidence, and security disabling.
  • •Exit-code masking and hollowed verifier detection catches silent neutering.
  • •Blocks and retries instead of just warning—forces agent accountability.
Weaknesses
  • •Claude Code-specific; won't work with other agent frameworks without rewriting.
  • •Judges agent against its own record, not ground truth—limited scope.
Category
AI/ML
Target Audience

Developers building or auditing AI coding agents

Similar To

LangChain validators · Agent monitoring tools

Similar Projects

Security●●●Banger

ClawDiary – Audit log and human approval gate for AI agents (Claw, MCP

Human-in-the-loop approval for AI agents via Telegram before risky ops.

Big BrainShip It
jetywolf
204mo ago
Security●●Solid

Clawsec - Open-source plugin for OpenClaw that blocks dangerous actions

Every tool call is caught by middleware, scored against built-in rulesets like destructive-commands, secrets/* and exfiltration/* in under 5ms, then enforced as block/confirm/allow via a clawsec.yaml — neat, pragmatic attack surface reduction. The demo and auto-generated config make onboarding trivial, but it currently reads as an OpenClaw-first solution; broader agent-framework integrations or stronger isolation guarantees would make this a must-install.

Solve My ProblemNiche GemSlick
subho007
114mo ago