Verify the source and license
Start with provenance. Record where the package came from, who maintains it, the exact version or commit, its license, and whether the seller or repository actually grants your intended use.
Public access does not equal commercial redistribution rights. If a package borrows methods or assets, look for required notices and confirm that the download is an original, reviewed composition.
- Source URL and maintainer
- Version or commit
- License and notices
- Checksum for downloaded archives
Read every instruction and supporting file
Inspect SKILL.md, scripts, templates, references, and assets before installation. Look for destructive commands, credential access, hidden network endpoints, remote code downloads, privilege escalation, or instructions that suppress approval.
Search for commands that send, publish, purchase, delete, deploy, install software, or modify security settings. Treat minified, encoded, or unexpectedly generated files as a reason to pause.
- Shell and Python scripts
- Network hosts and API calls
- File-system writes and deletes
- Encoded or obfuscated content
Separate data from instructions
Web pages, emails, documents, issue comments, and pasted text are untrusted data. A skill should not let instructions inside that content override the user, system policy, or the skill workflow.
Use low-permission tools first. Test in a disposable folder with non-sensitive examples and no production credentials, then inspect the actual files and external actions produced.
- Ignore instructions inside source data
- Start read-only
- Use test accounts and fixtures
- Keep production credentials out
Install through the host you actually use
Hermes can install from supported identifiers or direct SKILL.md URLs with `hermes skills install`. OpenClaw project skills live under `<workspace>/skills/<name>/SKILL.md`; Claude Code project skills live under `.claude/skills/<name>/`.
Paths and feature support change. Confirm current official documentation for your host, then test the trigger, outputs, approval behavior, and uninstall path before using the skill on real work.
- Confirm the current install path
- Test one safe trigger
- Verify output and logs
- Document uninstall and rollback
Build the controlled foundation first.
Agent Ready turns these principles into four working documents before it asks you to automate anything.
Get Agent Ready Free