Publish Guard

Publish Guard is a public ClawHub pre-release audit skill. Use it when the user says "publish guard", "release audit", "pre-release check", or wants to revie...

Security Scan

Publish Guard's code and runtime instructions are coherent with its stated purpose (pre-release public-surface audits); it reads repository files locally, requires Python, and does not request credentials or perform network exfiltration.

Purpose & Capability

Name/description match included artifacts: four Python scripts implement leak scanning, public-surface checks, README scoring, and audit rendering. The declared dependency (python3/python) is appropriate and proportional to the task.

Instruction Scope

SKILL.md instructs the agent to run the bundled Python scripts against a repo root. The scripts intentionally read repository files (README.md, SKILL.md, openai.yaml, .md docs, typical text files and tracked filenames) and search for secret-shaped strings and internal-language patterns. Reading repository files is expected for an audit, but the skill will scan any file in the provided repo tree (including .env-like files) so users should not point it at a workspace containing unrelated secrets.

Install Mechanism

No install spec; this is an instruction-only skill with bundled scripts. No external downloads or package installs are requested, so there is no high-risk installation behavior.

Credentials

The skill requests no environment variables or credentials. The scripts contain regexes that recognize secret patterns (OpenAI keys, AWS keys, etc.) for detection purposes — this is appropriate to a leak scanner and does not imply the skill needs or exfiltrates secrets.

Persistence & Privilege

always is false and the skill is user-invocable. It does not modify other skills or system-wide config. Autonomous invocation is allowed (platform default) but is not combined with broad credential requests here.

Assessment

Publish Guard appears to do what it says: it runs local Python scripts to detect leak patterns and audit public-facing docs. Before enabling or running it, consider: 1) it will read all scannable files under whatever repository path you give it (including README.md, SKILL.md, .env-like files and other text files), so do not point it at a workspace that contains unrelated secrets or private repos; 2) it does not request credentials or perform remote calls in the bundled scripts, and the code performs redaction of sensitive-looking snippets, but filenames and metadata may still appear in outputs; 3) confirm Python is available (python3 or python), and review the included scripts yourself if you have concerns; and 4) if you allow autonomous invocation, be aware the agent could run these scans without an explicit prompt — restrict invocation or test in a sandboxed workspace if you want an extra safety margin.

Search intent: publish guard, public surface review, release audit, pre-release check, clawhub publish audit

Goal

Audit the public surface before release:

• README
• SKILL.md
• agent metadata
• launch docs in the working tree
• obvious leak patterns

The output should answer one question clearly: publish now, or fix specific items first.

Use This Skill When

• the user wants to publish a GitHub repo or ClawHub skill
• the user wants a public-surface audit before a launch
• the repo may contain internal launch docs, secret-shaped strings, or operator-only wording
• the README or SKILL.md feels too insider-heavy or too long
• the first-run path may be broken, vague, or buried

Quick Start

1. Scan for obvious leak patterns.
   • python3 {baseDir}/scripts/scan_leaks.py --root <repo> --out <json>
2. Scan the public surface for audience-fit problems.
   • python3 {baseDir}/scripts/scan_public_surface.py --root <repo> --out <json>
3. Score the README or primary landing page.
   • python3 {baseDir}/scripts/score_launch_copy.py --readme <repo>/README.md --out <json>
4. Render one decision-ready audit.
   • python3 {baseDir}/scripts/render_public_audit.py --repo <repo> --leaks <json> --surface <json> --copy <json> --out <md>

Operating Rules

• Treat README.md, SKILL.md, agents/openai.yaml, and launch docs in the working tree as public.
• Public copy should describe the user job before it explains the internal theory.
• A public quick start should appear near the top and should be runnable without hidden context.
• Keep public default prompts short. Move deeper operating rules into the skill body or scripts.
• Flag internal launch docs in the repo unless they are intentionally private and excluded from the public package.
• Prefer a small set of concrete findings over a broad essay.

What To Flag

• absolute filesystem paths
• localhost URLs and websocket endpoints
• token-shaped strings and credential-like URLs
• missing or buried quick starts
• giant inventory sections or excessively long SKILL.md files
• public prompts that read like internal control instructions
• README intros that compare the repo to five other projects before explaining what it does

Bundled Scripts

• scripts/scan_leaks.py
  • Search the repo for obvious leak patterns and secret-shaped strings.
• scripts/scan_public_surface.py
  • Inspect README, SKILL.md, launch docs in the working tree, and public metadata for audience-fit issues.
• scripts/score_launch_copy.py
  • Produce a simple launch-copy score for the primary README.
• scripts/render_public_audit.py
  • Merge the JSON outputs into one concise markdown audit.