vibecheck
Audit AI-generated code before shipping. Static + AST scan for hardcoded secrets, SQL injection, eval/exec, shell injection, unsafe deserialization and TLS issues, with severity-capped 0-100 scoring and SARIF/JSON/Markdown reports. Optional autofix - every current fix changes behavior and requires the explicit --fix-unsafe opt-in; baseline export for legacy code. Zero-dependency Python CLI.
vibecheck v1.0.11
Static security and quality audit for AI-generated code. Scores a codebase 0-100, reports findings with evidence, and stays useful in CI. Zero dependencies, Python 3.8+.
Install
clawhub install thiennhant95/vibecheck
When to use
- The user asks for a code review, a security check, or whether code is safe to ship.
- Code was just generated by an AI assistant and needs a fast pre-merge audit.
- A Markdown, JSON, or SARIF report is needed for a PR or CI pipeline.
Usage
The tool lives at vibecheck.py inside this skill directory.
python3 <skill_dir>/vibecheck.py <path> # text output
python3 <skill_dir>/vibecheck.py <path> --format md # Markdown report
python3 <skill_dir>/vibecheck.py <path> --format json # JSON
python3 <skill_dir>/vibecheck.py <path> --format sarif # GitHub code scanning
python3 <skill_dir>/vibecheck.py <path> --min-score 80 --fail-on high # CI mode
python3 <skill_dir>/vibecheck.py <path> --fix-dry-run --fix-unsafe # preview autofix
python3 <skill_dir>/vibecheck.py <path> --fix --fix-unsafe # apply autofix (opt-in)
<path>: file or directory to scan (default.).--exclude PATTERN: regex path exclusion (repeatable).--baseline FILE/--write-baseline FILE: adopt legacy code and fail only on new findings.- Exit codes:
0pass,1fail,2bad input.
Verdicts and scoring caps
| Score | Verdict |
|---|---|
| 90-100 | CLEAN - production ready |
| 70-89 | OK - minor cleanup needed |
| 50-69 | RISKY - fix before merge |
| below 50 | DANGER - do not ship |
Open findings cap the score: HIGH caps it at 69 (RISKY) and CRITICAL at 30 (DANGER). A disabled TLS check or SQL injection can never be reported as production ready.
Permissions and scope
file_read: reads the files and directories being scanned.file_write: only when--fix,--fix-unsafe, or--write-baselineis used. All changes stay inside the scanned root; symlinks and paths outside it are skipped.env: reads process environment variables only; the tool makes no network calls and never executes scanned code.
Suppression
- Inline:
# vibecheck: disable=<rule-id>on the same line or the line above;# vibecheck: ignoreon the first line skips the file. - Project-wide:
.vibecheckignorewith one regex per line (POSIX paths on every OS).
Autofix
Every current autofix changes behavior, so it requires the explicit --fix-unsafe opt-in. Use --fix-dry-run first and review the diff. Autofix never edits outside the scanned root.
Reporting guidance
- Answer in the user's language and summarize findings by severity.
- For
critical/highfindings, explain the concrete risk and a fix; do not just quote the report. - Do not modify code unless the user asks; suggest fixes first.
- Reports may contain source context. Credential values are redacted, but treat reports as internal documents.
Limits
- Static analysis with one-hop alias/AST resolution; no cross-file data-flow.
- Unquoted
KEY=valuesecrets are checked only in.env*files; split secrets can evade. - Not a replacement for Semgrep, Bandit, or CodeQL; it is a fast pre-merge gate.
Related skills
vibegit— estimate AI-assisted work and slop signals from git history:clawhub install thiennhant95/vibegitskill-firewall— scan third-party agent skills before installing them:clawhub install thiennhant95/skill-firewallundo-first— reversible-action guard with snapshots, risk gate, and verified restore:clawhub install thiennhant95/undo-first
Source, tests, and benchmark: https://github.com/thiennhant95/vibecheck
License: MIT.
- Downloads
- Security audit
- Pass
- Last updated
- 3d ago
- Current version
- v1.0.11
- License
- MIT-0
