agentdryrun scan: dry run on your Claude Code history

    1

    See what your coding agent did that you would never have approved. Read-only, local, no network, no dependencies.

    Free

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    agentdryrun scan: dry run on your Claude Code history

    Example session with this skill installed

    /agentdryrun-scan --days 30

    • Read your context and instructions
    • Compiled the agentdryrun scan
    • Generated the document

    This is a run on a small invented demo project (not a real history), so you can see the shape of the output.

    Summary

    10 sessions and 35 tool calls in the last 30 days were judged against the starter rules pack:

    14 would pass, 16 would need approval, 5 would be refused. The tool output the model read was about 11,210 tokens, and 3 of the 35 calls failed.

    What the agent didCallsWould be refusedWould need approval
    Infrastructure or cloud deletes312
    Writes to settings, CI, startup files, system folders312
    Force pushes / remote branch deletes110
    Destructive database commands110
    Files or output sent to a remote host110
    Recursive deletes403
    MCP tools that do not look read-only302
    Work thrown away (reset --hard, clean, ...)202

    The scanner wrote one local HTML report card to ~/agent-report-card.html, without command text. Nothing was blocked, changed or sent.

    Next steps

    • Open the report card and look at the refused rows first.
    • The report shows project folder names but no command text, so it is safer to share than a full report.
    • This is pattern matching: treat a count as a question to check, not as a verdict.

    agent-report-card.html

    PDF · document

    Generated

    Example file from a real run - the skill writes it into your workspace.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    One command reads the session history Claude Code already keeps on your machine and shows what your rules would have stopped: recursive deletes, force pushes, curl | sh, secret files read, uploads, destructive database and cloud commands. It also shows where tool output piled up. It writes one local HTML report card. Nothing is sent anywhere and nothing is changed.

    What it reads and writes: your Claude Code transcripts in ~/.claude/projects (or the folder named in CLAUDE_CONFIG_DIR), and one report, agent-report-card.html, in your home folder. The report is written without command text (--no-previews); it still shows project folder names. The scanner itself makes no network connection; the short summary it prints appears in your Claude Code session like any command output. Words such as .ssh or api_key appear in its rule patterns because it looks for risky commands.

    Honest limits: it is pattern matching on commands, so it misses things and flags a few harmless ones. It reads Claude Code transcripts only. It is not a sandbox and stops nothing. Needs Python 3.9+.

    Open source scanner (Apache-2.0): github.com/ntinosgkiou-code/agentdryrun

    Requires Python 3.9+. Only scans Claude Code transcripts, not other agents. Pattern matching may miss obfuscated commands or runtime-generated paths. Does not function as a real-time sandbox.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 15 days ago

    • Free with an Agensi account

    Requires

    Python

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Environment Variables

    File Scopes

    ~/.claude/projects/**/*.jsonl
    ~/agent-report-card.html

    Reads only the Claude Code transcripts in ~/.claude/projects (or the folder named in the CLAUDE_CONFIG_DIR environment variable) and writes one report, ~/agent-report-card.html. It makes no network connection: urllib is imported only for urllib.parse, which splits web addresses that appear inside commands. Words such as .ssh or api_key in the code are rule patterns used to detect risky commands; the scanner never opens those files. It starts no other program and installs nothing.

    Listed15 days ago

    What's inside

    Frequently Asked Questions

    The skill reads local transcript logs stored by the Claude Code CLI. It identifies high-risk commands, such as recursive deletes or force pushes, and flags access to sensitive files like .env or credentials that may have occurred during automated sessions.

    This skill is specifically designed to analyze session history from the Claude Code CLI. While it requires Python 3.9+ to run the analysis, it does not support other AI agents or generic shell histories at this time.

    The scanner itself makes no network connection: it reads the transcripts on your machine and writes one local report. But it runs inside your Claude Code session, so the short summary it prints (counts and kinds of action) appears in that session like any command output, and your model provider sees it as part of the conversation. The report file is written without command text and is not opened by the agent.

    You will receive the Python-based scanning tool, a starter pack of security rules for command categorization, and the report generation module. Setup is straightforward, requiring only a local Python environment and access to your Claude Code project directory.

    The skill can detect direct commands like 'rm -rf' or 'git push --force' recorded in the transcripts. However, it cannot 'see' the internal logic of custom scripts the agent might have written and executed, nor can it detect risks involving paths generated dynamically at runtime.

    New in Code Quality & Review