Hermes Skill Publishing Guide

Guide to publishing and sharing Hermes skills through ClawHub, including safe login practices, release hygiene, and security scan preparation.

Security Scan

This is an instruction-only publishing guide whose commands and recommendations match its stated purpose; nothing in the file requests unrelated credentials or installs arbitrary code.

Purpose & Capability

The name/description (publishing/share guide for ClawHub) aligns with the runtime instructions: login, publish, inspect, delete, merge, and a security checklist. There are no unrelated binaries, services, or unexplained permissions requested.

Instruction Scope

SKILL.md contains only publishing-related guidance and example clawhub commands and explicitly advises safe handling of tokens and removal of secrets. It does include destructive actions (publish/delete/merge) which are expected for a publishing guide, but it does not instruct the agent to read unrelated files, exfiltrate data, or contact unexpected endpoints.

Install Mechanism

No install spec or code files are present; this instruction-only format is low-risk because nothing will be written to disk by an installer.

Credentials

The document demonstrates reading a CLAWHUB_TOKEN from stdin and unsetting it, but the registry metadata lists no required env vars. This is a minor mismatch but reasonable: the token is optional/interactive in examples. Users should treat CLAWHUB_TOKEN as a sensitive secret and not expose it in history or logs.

Persistence & Privilege

The skill does not request persistent/always-on presence (always:false). It is user-invocable and the instructions show commands that can modify or delete skills — appropriate for a publishing guide. Note: because agent invocation is allowed by default, avoid letting the agent run these commands autonomously unless you trust it to act only under explicit user direction.

Assessment

This guide appears coherent and safe: it gives expected clawhub commands and sensible security advice. Before following examples, double-check slugs/owner names to avoid accidental delete/merge, provide CLAWHUB_TOKEN interactively (as shown) rather than embedding it, and review any command before running it. Since the file is instruction-only, no code will be installed, but be cautious about allowing autonomous agent execution of publish/delete commands — run them manually or confirm prompts to prevent accidental changes.

When to use

• You are preparing a skill for ClawHub publication.
• You need to update a published skill version.
• You want a short checklist for passing the security scan.

Safe login

Prefer reading the token privately instead of pasting it directly into a shared command line history.

read -s CLAWHUB_TOKEN
clawhub login --token "$CLAWHUB_TOKEN" --no-browser
unset CLAWHUB_TOKEN

Publish a skill

clawhub publish /path/to/skill-folder \
  --slug "skill-slug" \
  --name "Display Name" \
  --version "1.0.0" \
  --changelog "What changed in this release" \
  --tags "latest,tag1,tag2"

Useful commands

• clawhub whoami
• clawhub inspect <slug>
• clawhub search <query>
• clawhub delete <slug>
• clawhub skill merge <source-slug> <target-slug>

Security scan checklist

Before publishing:

• Remove secrets, tokens, and internal URLs.
• Remove fake commands or instructions for tools that are not actually included.
• Use official download sources only.
• Avoid hardcoded system paths and unsafe command examples.
• Keep examples focused on the skill's real behavior.
• Rewrite or remove anything that could look like exfiltration, persistence, or privilege escalation.

Release hygiene

• Bump the version every time you republish.
• Keep the changelog factual and short.
• Use English titles and descriptions if the target audience is broad.
• Merge or delete duplicate skills so only the canonical version remains active.