Coding-agent monitoring · Claude Code

    Claude Code monitoring for teams

    Claude Code monitoring means reading every Claude Code session your team runs to find what keeps going wrong: the same failure again, instructions retyped by hand, CLAUDE.md rules nobody follows, secrets in prompts. Hervé does it across the whole team, fixes the cause in CLAUDE.md and skills, and checks the next sessions.

    Last updated

    Why one Claude Code session can't show you the problem

    Claude Code reads CLAUDE.md, the rules in .claude/rules/ and the skills in .claude/skills/ to know how your team works. In its own docs: “CLAUDE.md files are markdown files that give Claude persistent instructions for a project, your personal workflow, or your entire organization. You write these files in plain text; Claude reads them at the start of every session.”[1]

    Each developer only sees their own sessions. Nobody sees that three people hit the same wall last week, that the same instruction gets typed by hand every morning, or that a rule in CLAUDE.md has not changed what the agent does in a month. 84% of developers use or plan to use AI tools in their development process, up from 76% a year earlier.[3] The more of the team runs Claude Code, the bigger that blind spot gets.

    What Hervé finds in Claude Code sessions

    Hervé reads every captured session across the team and groups what keeps going wrong into findings, each with how many sessions and developers it touched:

    • Recurring failures: the same mistake across sessions and developers, with the human time or tokens it cost.
    • Repeated instructions: guidance team members retype by hand that belongs in one shared rule or skill.
    • Unused instructions: sections of CLAUDE.md and skills that no session follows or uses, but that still take up context.
    • Leaks: a secret pasted into a prompt, or printed by the agent. GitGuardian measured a 3.2% secret-leak rate in Claude Code-assisted commits, against a 1.5% baseline across all public GitHub commits.[2]

    How Hervé fixes the cause at the source

    For each finding, Hervé proposes the fix where Claude Code will read it next time: a change to CLAUDE.md, the rules in .claude/rules/ and the skills in .claude/skills/. The fix is opened as a pull request on your repository, so it goes through your usual merge flow and reaches the whole team, not one developer's private memory.

    How Hervé guards the next Claude Code sessions

    Once a fix is merged, Hervé checks every new session against your rules and compares the before and after, so you see whether the problem stopped. You get a weekly update, and you stay in control of what the agents are told.

    How to set up Claude Code monitoring

    1. Install the rv CLI. Follow the one-minute setup for macOS, Linux or Windows.
    2. Enable capture for Claude Code. In each repository, run:
      my_repo $ rv install --agent claude-code

      Sessions are saved to a branch of your own repository, and the last 30 days of local sessions are backfilled by default.

    3. Have every team member do the same once. Hervé only sees the sessions it captures, so each developer runs rv install once. The first findings come in week 2.

    Common questions

    What does Hervé read in a Claude Code session?

    The prompts, the steps the agent took and the files it touched, from the branch rv install writes to your own repository. Hervé reads sessions and code on demand: transcripts, diffs and file contents never go into our database. We keep findings, counts and file names.

    Can't I just ask Claude Code to clean up CLAUDE.md?

    You can, for one session. Claude Code sees what happened in front of it; it can't see that three developers hit the same wall last week, or which rules no session has followed in a month. Hervé reads every session across the team, then checks whether the fix held.

    Which files does Hervé change to fix a Claude Code problem?

    The ones Claude Code reads its instructions from: CLAUDE.md, .claude/rules/ and .claude/skills/. Each fix is opened as a pull request on your repository, so it goes through your usual merge flow and reaches the whole team.

    Do we have to change how we use Claude Code?

    No. Each team member runs rv install once. Everyone keeps using Claude Code, GitHub or GitLab, and your merge flow exactly as they are.

    What if the team uses more than one coding agent?

    Hervé captures sessions from Claude Code, Cursor, GitHub Copilot CLI, OpenAI Codex, and OpenCode, so a team that mixes agents is covered by the same install.

    How do we start?

    With a 6-week paid pilot for your whole team, fully refunded if we find nothing actionable. Talk to a founder to set it up; the pilot details and price are on the pricing page.

    See pilot details & pricing →

    Sources

    1. How Claude remembers your project, Anthropic.
    2. The State of Secrets Sprawl 2026, GitGuardian.
    3. 2025 Developer Survey: AI, Stack Overflow.

    Monitoring other coding agents

    See what your team's Claude Code sessions are hiding.

    Talk to a founder

    6-week paid pilot, fully refunded if we find nothing actionable.