메뉴
A deliberately leaky demo skill used to show what skill-sanitizer catches before you publish. Do not ship this.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Pre-share leak scanner for Claude Code skills (or any folder). Scans BEFORE you publish for secrets (API keys, tokens, private keys, .env values, high-entropy strings), PII (emails/phones), local machine paths (C:\Users\<name>, /home/<name>, /Users/<name>, UNC, internal hostnames), and client/proper-noun names from a LOCAL private dictionary; also flags service-specific code (jira/github/gitlab/slack/salesforce/notion/linear/ stripe/aws/google) that should be split into a dedicated service plugin. Use when the user says "sanitize this skill", "before sharing a skill", "check for leaks", "is this skill clean", "is this safe to commit/publish/ open-source", "scrub secrets", "redact before commit", or "pre-push leak check". Cross-platform, stdlib-only Python, fail-closed.
Package an AI skill, prompt workflow, or internal operating habit into a public-ready release with a catchy hook, safe redaction, useful examples, demo tasks, README copy, and launch messaging. Use when preparing a skill for GitHub, social sharing, marketplace submission, documentation, or community feedback.
Lightweight multi-agent research orchestration using one coordinator, three domain leads, and three lightweight research agents. Use for medium-size research, market scans, competitor comparisons, prior-art style exploration, report planning, and any task that benefits from parallel domain decomposition without running a full high-cost research lab.
Analyze folders of Markdown/text research notes, Obsidian vaults, paper cards, invention notes, or memory exports as a practical knowledge map. Use when asked to find center topics, noisy or low-signal notes, possible contamination, overgrown topic groups, orphan notes, cleanup actions, or a useful public knowledge-hygiene workflow.
| name | dirty-skill |
| description | A deliberately leaky demo skill used to show what skill-sanitizer catches before you publish. Do not ship this. |
This folder is a teaching fixture for skill-sanitizer. It contains planted,
clearly-fake-but-realistic leaks so you can watch the scanner BLOCK on a publish
attempt. Nothing here is a real credential — the values are dummies sized to
look real so the scanner treats them seriously.
| Class | Where | Planted value (fake) |
|---|---|---|
| SECRET | sync.py:5 | a ghp_ GitHub token (36 hex-ish chars) → hard BLOCK |
| PROPRIETARY | sync.py:8 | a hardcoded home-dir path with a username → WARN |
| PII | sync.py:11 | a personal gmail.com email → WARN |
| SERVICE CODE | sync.py | an internal Jira marker (.atlassian.net, customfield_*) → plugin-split advice |
cd examples/dirty-skill
python ../../scanner.py .
Expected: a report with BLOCK=2 WARN=4 and exit code 1. The full expected
report is in ../EXPECTED_OUTPUT.md. Fixing the four
planted leaks (env-var the token, swap the path for ~, use an example.com
email, move the Jira code to a plugin) takes the same folder to exit 0.