Snitch: deterministic claim verifier for AI agent transcripts
Snitch is a tool that watches AI agent transcript files (Cursor, Claude Code, Codex, Pi, OpenCode) and verifies claims made in prose against actual evidence like tool calls, shell output, filesystem changes, git history, and session context. It uses deterministic regex patterns to extract claims and cross-references them, flagging inconsistencies. It helps developers trust their coding agents by catching when an agent's description doesn't match reality.
Entities
Related
Confessor: Analyzes Claude Code transcripts to show what the AI agent did on your computer
Confessor is an open-source tool that reads the JSONL log files Claude Code leaves on disk and generates a human-readable report of every action the AI agent performed, including tool calls, file reads, and shell commands. It helps developers understand and audit the behavior of AI coding agents on their machines.
agentsweep CLI scans AI coding agent history files for leaked secrets
A new open-source CLI tool called agentsweep scans history files of AI coding agents (Codex, Cursor, Claude Code, Cline, Aider) for plaintext secrets like API keys, DB URLs, and crypto seed phrases. It uses ~191 detection rules from gitleaks plus a dedicated BIP-39 seed phrase detector, addressing a security risk where pasted secrets persist in agent context and can be re-exposed.
CodeInspectus: open-source security scanner for AI-generated code
CodeInspectus is a fully open-source, local security scanner that checks AI-generated code for vulnerabilities. It covers 32 checks (13 AI-specific + 19 SAST) and 200+ secret/API-key patterns, catching issues like hardcoded secrets in client-side code, exposed API keys, and insecure RLS policies. It helps developers secure projects built with LLM-generated code.
Git-aware AI debugger that checks out old commits to fix production bugs
A tool that makes AI coding assistants (like Cursor or Claude Code) automatically checkout the git commit corresponding to a production error before debugging, preventing the agent from analyzing current code that has shifted. It solves the problem of AI agents hallucinating fixes because they look at the present state of files while the bug existed in a past commit.
HarnessTrim: a deterministic, benchmarked token-economy layer across Claude Code, Codex & OpenCode
HarnessTrim is a new tool that provides a deterministic, benchmarked token-economy layer across coding agent harnesses like Claude Code, Codex, and OpenCode. It uses idempotent reducers to cut token waste from noisy tool output, model verbosity, thinking tokens, and instruction files, without involving an LLM in the reduction path. The tool coordinates existing single-channel solutions (e.g., Caveman, RTK) behind a unified cross-harness policy, and is cache-aware to avoid touching cacheable prompts.
