# AI use cases for Reverify (2akouwu/reverify) Repo: [2akouwu/reverify](https://github.com/2akouwu/reverify) | Use cases: 4 | Page: https://dedupedai.com/use-cases/2akouwu-reverify/ Reverify is a Python toolkit and MCP server where deterministic tools verify every AI claim about a binary or code before it counts as fact. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - Check an AI's claims about a binary before trusting them: https://dedupedai.com/use-cases/2akouwu-reverify/verify-ai-claims-about-binary/ - Test an AI's code rewrite against the original before merging: https://dedupedai.com/use-cases/2akouwu-reverify/gate-ai-rewrite-with-equivalence-tests/ - Stop a long AI coding session from losing grounded facts: https://dedupedai.com/use-cases/2akouwu-reverify/keep-long-agent-session-from-drifting/ - Let a coding agent call Reverify's checks through MCP: https://dedupedai.com/use-cases/2akouwu-reverify/give-agent-verifier-tools-over-mcp/ ## How to check an AI's claims about a binary before trusting them with Reverify Repo: [2akouwu/reverify](https://github.com/2akouwu/reverify) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README Turn an AI's guesses about a binary into VERIFIED or REFUTED verdicts with evidence. **Who it is for:** A reverse engineer or security analyst who has asked an AI to describe a function, struct, or offset in a binary and needs to know which claims are real. **The problem:** Language models confidently invent offsets, sizes, and function behavior when reading binaries, and there is no easy way to tell which statements are true. **The outcome:** Each claim is returned as VERIFIED, REFUTED, or INCONCLUSIVE with the bytes the tools observed, and only verified claims are kept. ### Steps 1. Run `pip install reverify` to install the CLI and MCP server. 2. Run `reverify auto sample.bin --json` to get the tool-derived view of the file. 3. Write each AI claim as a JSON object, for example kind instructions with an offset and a list of mnemonics. 4. Run `reverify verify sample.bin --claim '{"kind": "instructions", "offset": 4096, "mnemonics": ["push", "mov", "sub"], "note": "function prologue"}'` and read the verdict. (inferred, check the README) 5. For many claims, save them to claims.json and run Reverify verify with --claims-file, then check the exit code, which is non-zero if anything is refuted. ### Prompt to paste into your agent ```text Using the reverify repo at 2akouwu/reverify, check each claim you made about with reverify verify and its --claim JSON format. Report each claim as VERIFIED, REFUTED, or INCONCLUSIVE with the evidence bytes, and do not state any claim as fact unless the tool returned VERIFIED. ``` ### Caveat The README reports its benchmark numbers on Windows system files, so accuracy on other binaries is not promised by the README. Risk, reverse engineering: The README says Reverify is for authorized reverse engineering only, such as malware analysis, CTF, interoperability research, and software you own or are permitted to analyze. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/2akouwu-reverify/verify-ai-claims-about-binary/ ## How to test an AI's code rewrite against the original before merging with Reverify Repo: [2akouwu/reverify](https://github.com/2akouwu/reverify) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README Run a candidate and a reference implementation on shared inputs and catch any disagreement. **Who it is for:** A developer who asked an AI agent to refactor or rewrite a Python or C function and wants proof it still behaves the same. **The problem:** An AI refactor can look correct while changing behavior on edge cases, and reading the diff does not prove equivalence. **The outcome:** A refuted result comes back with the input and both outputs, so the developer can see exactly where the rewrite differs. ### Steps 1. Run `pip install reverify` to install the tool. 2. Prepare the reference implementation and the AI's candidate implementation as separate files or functions that Reverify equiv can load. (inferred, check the README) 3. Run `reverify equiv --lang python` and supply the candidate and reference paths as the README's equiv usage requires. (inferred, check the README) 4. Read the result: a refutation shows the input and both outputs that disagree. 5. Fix the candidate and rerun until the check passes before merging. ### Prompt to paste into your agent ```text Using the reverify repo at 2akouwu/reverify, run its equiv check with --lang python to compare the AI-rewritten against the original reference over shared inputs. Report the result and, if refuted, the input and both outputs, and do not accept the rewrite unless the check agrees. ``` ### Caveat The README gives the equiv command name and language flag but not the exact argument format, so the input setup steps here are inferred and need checking against the tool's own help. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/2akouwu-reverify/gate-ai-rewrite-with-equivalence-tests/ ## How to stop a long AI coding session from losing grounded facts with Reverify Repo: [2akouwu/reverify](https://github.com/2akouwu/reverify) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README Hand off a long task to a file-backed state so a fresh context keeps verified facts. **Who it is for:** A developer running a long analysis task in Claude Code, Codex, or Gemini CLI whose context keeps degrading after auto-compaction or repeated clears. **The problem:** Auto-summaries drop details and let unverified guesses carry forward, so long tasks drift and need restarts. **The outcome:** Verified and refuted results are saved to a ledger on disk, and a new session resumes from the same grounded position without re-proposing known false claims. ### Steps 1. Run `pip install reverify` to install the tool. 2. Run `reverify rollover` to hand the session off to a file and start a fresh one. 3. Run `reverify ledger target.exe` to see what is established and what is known false. 4. Run `reverify ledger --hook` to register the Claude Code SessionStart hook for compact, clear, and resume events. ### Prompt to paste into your agent ```text Using the reverify repo at 2akouwu/reverify, run reverify rollover to hand off the current session, then run reverify ledger and confirm the established and known-false facts appear in the new session before continuing. ``` ### Caveat The README says the rollover works in Claude Code, Codex, Gemini CLI and OpenCode, but the hook command is documented only for Claude Code, so other agents need their own setup. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/2akouwu-reverify/keep-long-agent-session-from-drifting/ ## How to let a coding agent call Reverify's checks through MCP with Reverify Repo: [2akouwu/reverify](https://github.com/2akouwu/reverify) | Difficulty: Advanced | Time: 45 minutes | Basis: From the README Connect Claude Code or Cursor to Reverify's MCP tools so the agent verifies its own claims. **Who it is for:** A developer using Claude Code or Cursor who wants the agent to check its binary-analysis claims with tools instead of trusting its own output. **The problem:** Agents assert facts about binaries in chat with no check, and they lose verified state when their context is reset. **The outcome:** The agent calls re_verify_claim for each claim and re_ledger after a reset, so only grounded results are reused. ### Steps 1. Run `pip install reverify` to install the CLI and MCP server. 2. Add the Reverify MCP server to your agent's MCP configuration, following your agent's MCP setup steps. (inferred, check the README) 3. Ask the agent to verify each binary claim through re_verify_claim. 4. After the host compacts or clears context, ask the agent to call re_ledger and continue from the returned facts. ### Prompt to paste into your agent ```text Using the reverify MCP server from the repo 2akouwu/reverify, verify each claim about with re_verify_claim and recover prior grounded facts with re_ledger after any context reset. Report which claims were VERIFIED and which were REFUTED. ``` ### Caveat The README names the MCP server and its tools but does not give the exact MCP configuration for each agent, so the setup step is inferred. Risk, reverse engineering: Use the MCP tools only on binaries you are authorized to analyze, as the README requires. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/2akouwu-reverify/give-agent-verifier-tools-over-mcp/