Is your Claude Code workflow dangerous for production?

Generate a local security report showing whether your Claude Code behavior is safe, risky, or dangerous in production.

Track the patterns that make Claude Code usage risky.

cc-vibeguard parses transcript and tool activity locally on your device, groups it into concrete security risk categories, and shows where Claude Code is reading secrets, running dangerous commands, bypassing permissions, or operating with too much autonomy.

Secrets exposure

See when secret handling becomes a habit, not a one-off.

The report measures exposed keys, .env reads, credential patterns, and repeated secret handling across projects so risky behavior shows up as a pattern, not an anecdote.

  • Counts secret types and repeated exposures across projects
  • Separates reads, writes, and environment file touch points
  • Highlights the bad habits and hotspots that deserve attention

secret exposure

14/100

KEY= values264
SECRET= values205
Bearer tokens54

reads

225

writes

55

projects

10

Remote access and commands

Measure when Claude starts acting too close to production.

Vibeguard tracks SSH usage, remote command patterns, and destructive shell behavior so you can see where assistance turns into risky operational behavior.

  • Separates interactive SSH from remote command execution
  • Groups destructive commands by severity and category
  • Calls out force pushes, resets, and other high-risk patterns

remote and commands

1,031 ssh

Critical18
High37
Medium64
Low91

top host

prod-host-01

remote user

ops-user@remote-a

internal box

internal-box-03

Permission discipline

Show where speed is winning over guardrails.

Bypass mode, approval patterns, autonomous agents, and human overrides are stitched into one view so you can measure how supervised the workflow really is.

  • Tracks bypass vs default permission modes over time
  • Surfaces agent spawning and oversight behavior
  • Counts denials, interrupts, and destructive catches

permission discipline

19% bypass

Bypass

19%

Accept edits

44%

Default

37%

agents

164

human catches

31

elevated autonomy

88 spawned with more freedom

Ranked output

End with a report people can actually read and compare.

The final report condenses raw behavior into an overall score, risk priorities, and a developer ranking that is easy to review, discuss, and share inside a team.

  • Builds a weighted overall safety score
  • Summarizes the biggest risks and what is working well
  • Generates a clear report that can be shared internally

share report

final rank

The Boundary Pusher

Fast, productive, and a little too comfortable around production-shaped edges.

Secrets

14

Remote

56

Commands

23

Oversight

30

PRIVACY AND LOCAL

The scan stays local. The report is yours.

cc-vibeguard reads local Claude Code transcript history, calculates the metrics on your machine, and writes an HTML report you can inspect or share. Nothing ever leaves your device.

No transcripts uploaded

Zero network calls

Your prompts, code, and project contents never touch a hosted dashboard.

Reads transcripts in place

Parses Claude Code history directly from disk and keeps every byte there.

HTML report output

Readable and portable

One self-contained HTML file. Open it in any browser, no server required.

Shareable on your terms

Inspect raw metrics, share the file with a teammate, or archive it for later review.

Built for peer review

Self-audit or team check-in

Run it on yourself, compare with a teammate, or use inside a team to spot shared bad habits.

Measurable, not anecdotal

Hidden risk patterns become concrete numbers instead of hallway conversations.

Join vibenalytics

Want to measure the tokens, tools, and sessions behind every Claude Code project? Vibenalytics turns your day-to-day AI work into one live dashboard - usage, costs, patterns, all of it.

$curl -fsSL https://vibenalytics.dev/install.sh | bash

Privacy-first analytics and security awareness for AI-assisted development.