Oathe Security Badge

oathe / scan / reports / safe

Is poshan0126/dotclaude safe?

Yes. poshan0126/dotclaude is safe to install. Oathe's behavioral security audit gave the dotclaude skill by poshan0126 a trust score of 91/100 with 5 findings, none critical or high. Report updated

https://github.com/poshan0126/dotclaude

91
SAFE

Is poshan0126/dotclaude safe to install?

dotclaude is a legitimate Claude Code configuration toolkit providing safety hooks and development workflow skills. No prompt injection, data exfiltration, or malicious code was detected. The hooks serve genuine defensive purposes (blocking force-push, secret commits, rm -rf, etc.) and are architecturally transparent. Minor concerns include the large hook execution surface on every tool use, the test-writer skill auto-triggering without explicit invocation, and the hooks self-protecting from Claude modification once installed — all design choices the user should understand before installing.

What security issues were found in poshan0126/dotclaude?

Category Scores

Prompt Injection 92/100 · 30%
Data Exfiltration 93/100 · 25%
Code Execution 85/100 · 20%
Clone Behavior 95/100 · 10%
Canary Integrity 100/100 · 10%
Behavioral Reasoning 88/100 · 5%

Findings (5)

LOW Hook scripts execute on every Edit/Write/Bash tool use -10 ▶

Eight shell scripts are wired as PreToolUse/PostToolUse hooks that run on every file edit and shell command. While the content is verifiably benign, this creates a large trusted execution surface. Any future supply-chain compromise of this repo would immediately affect all users on next git pull.

Evidence: plugins/safety-hooks/hooks/hooks.json wires protect-files.sh, warn-large-files.sh, scan-secrets.sh, block-dangerous-commands.sh as PreToolUse hooks with no content-hash pinning

LOW test-writer skill auto-triggers without explicit user consent -8 ▶

The test-writer SKILL.md does not set disable-model-invocation:true. The comment in the file explicitly notes this enables Claude to auto-trigger it after adding features. This means the skill injects substantial test-writing instructions into every session where Claude adds code, without the user explicitly invoking /test-writer.

Evidence: skills/test-writer/SKILL.md frontmatter: no disable-model-invocation field; comment reads '# No disable-model-invocation. Claude can auto-trigger this when adding features.'

LOW context-budget --api mode sends file content to Anthropic tokenizer endpoint -7 ▶

When invoked with --api, context-budget skill reads .claude/ configuration files and sends their content to api.anthropic.com/v1/messages/count_tokens. This is the official Anthropic API, the flag is opt-in, and the purpose is legitimate token counting. However, users should be aware their CLAUDE.md and rule content leaves the local machine.

Evidence: skills/context-budget/SKILL.md: curl -s https://api.anthropic.com/v1/messages/count_tokens with file content in request body

LOW Hooks self-protect from Claude modification -12 ▶

protect-files.sh blocks Claude from editing any file under .claude/hooks/, including itself. This means once installed, the hooks cannot be updated or removed by Claude — only by the user manually. This is a design choice that prioritizes integrity over flexibility, but users should understand they cannot ask Claude to modify or disable safety hooks.

Evidence: hooks/protect-files.sh: .claude/hooks/*|*/.claude/hooks/*) emit deny "Cannot edit hook scripts. These enforce security boundaries."

INFO Canary file accesses predate the clone and are attributable to audit infrastructure 0 ▶

The audit logs show accesses to .env, id_rsa, .aws/credentials, .npmrc, .docker/config.json at 15:06:47. The git clone began at 15:06:53. These accesses are from the audit system's own sudo pre-check process (pid 1102), not from any skill code. Canary integrity check confirms all files intact.

Evidence: auditd PATH events at timestamp 1794409607.933 for /home/oc-exec/.env etc; git clone EXECVE at 1794409613.458 — 6 seconds later

Should I install poshan0126/dotclaude?

Oathe's verdict for poshan0126/dotclaude is SAFE with a trust score of 91/100. Recommendation: Install.