OpenClaw Skill Review Checklist Before Install (2026)

If your team installs many community skills, speed matters. But skipping review creates fragile setups and avoidable incidents.

This checklist gives you a repeatable "go/no-go" gate before enabling any OpenClaw skill.

TL;DR

• Run canonical install and verify local SKILL.md before calling a skill "verified."
• Keep sourceDescription and originalSkillMd strictly source-derived.
• Require permission tiering, dry-run, and rollback triggers before production use.

Table of contents

• Who this guide is for
• 1. Verify canonical source first
• 2. Validate source fields exactly
• 3. Check permissions and blast radius
• 4. Require rollback triggers in advance
• 5. Run a dry run before real tasks
• 6. Record verification log for reproducibility
• Final go/no-go rule
• Pre-install review use-case mapping
• Verification log template
• Publish decision memo template
• Metrics snapshot
• Failure -> Fix example
• Limitations and scope
• Conclusion: pre-install decision rule
• FAQ
• Next steps
• References

Who this guide is for

• Teams installing third-party OpenClaw skills at high frequency
• Reviewers responsible for security and source integrity gates
• Editors maintaining verified skill pages with reproducible evidence

If a skill requires broad high-risk permissions, this checklist should be paired with a dedicated security review.

1. Verify canonical source first

Run installer first and treat the local installed path as source of truth.

npx -y skills add <repo-url> --skill <skill-name> -y -g
test -f ~/.agents/skills/<skill-name>/SKILL.md
ls -la ~/.agents/skills | rg "<skill-name>"

If canonical source is missing, do not publish as fully verified.

2. Validate source fields exactly

From canonical SKILL.md:

• sourceDescription must equal frontmatter description.
• originalSkillMd must contain full body with YAML frontmatter removed.

Do not replace source-derived fields with rewritten copy.

3. Check permissions and blast radius

Before install, classify permissions into:

• Low risk: file read workflows
• Medium risk: shell, web with clear limits
• High risk: broad shell/network plus destructive operations

If a skill needs broad power, require extra rollback controls.

4. Require rollback triggers in advance

A skill is not production-ready without explicit rollback conditions.

Minimum rollback triggers:

1. Functional regression in core workflows
2. Repeated runtime failures after configuration checks
3. Security review flags unresolved permission abuse risk

5. Run a dry run before real tasks

Never start with destructive tasks.

Prompt template:

Run a non-destructive dry run for this skill, list required permissions, and show fallback behavior if a dependency is missing.

6. Record verification log for reproducibility

Store checks with Pass / Partial / Fail so future editors can audit quickly.

Typical checks:

• Canonical install command executed
• Canonical file exists
• Frontmatter description extracted
• Full source body extracted

Final go/no-go rule

Ship a skill page only when:

• canonical source is available,
• source fields are exact,
• permissions are reviewed,
• rollback triggers are documented,
• dry run passes.

OpenClaw skill review checklist scoring

Use a simple scoring model to standardize go/no-go decisions:

Set a release threshold (for example 85/100) and require security review for any high-risk permission profile.

Pre-install review use-case mapping

Verification log template

Use this copyable log shape for each reviewed skill:

skill: <skill-name>
canonical_install:
  command: "npx -y skills add <repo-url> --skill <skill-name> -y -g"
  status: pass|partial|fail
source_checks:
  canonical_file_exists: pass|partial|fail
  source_description_exact: pass|partial|fail
  original_skill_md_extracted: pass|partial|fail
risk_review:
  permission_tier: low|medium|high
  rollback_triggers_defined: yes|no
dry_run:
  status: pass|partial|fail
reviewer: <name>
date: <YYYY-MM-DD>

Publish decision memo template

Use this short memo format before publishing a verified skill page:

skill_page: <url>
decision: approve|block|summary-only
primary_reason: <one sentence>
source_integrity_status: pass|partial|fail
permission_risk: low|medium|high
dry_run_status: pass|partial|fail
rollback_plan_ready: yes|no
reviewer: <name>
date: <YYYY-MM-DD>

Metrics snapshot

Method note: results reflect one internal adoption cycle (n=20 reviewed skills) and should be used as comparative guidance.

Failure -> Fix example

• Failure: page was published from a guessed source path and later found mismatched frontmatter description.
• Fix: enforce canonical install path checks and block publish until sourceDescription and originalSkillMd are extracted from local SKILL.md.

Limitations and scope

• This checklist reduces risk but does not replace full security review for high-privilege skills.
• Risk tiers here are generic; adapt thresholds to your own environment and policy.
• Fast-moving upstream repos can invalidate assumptions, so periodic re-verification is required.

Conclusion: pre-install decision rule

No canonical source proof means no verified publish. No least-privilege evidence means no production rollout. Treat this checklist as a release gate, not optional documentation.

FAQ

What is the minimum evidence required to publish as verified?

You need canonical install proof, local source file verification, exact source-derived field extraction, and a passed dry-run log.

Should summaries ever replace originalSkillMd content?

No. Summaries can complement analysis, but source-derived fields must remain exact and unmodified from canonical source content.

How often should we re-verify previously approved skills?

Re-verify on upstream updates, dependency changes, or at fixed intervals such as every 30 to 60 days for active skills.

When is summary-only mode the right choice?

Use summary-only when canonical source cannot be fetched or source frontmatter cannot be verified with confidence.

For related setup guidance, see:

• How to Install OpenClaw Skills
• OpenClaw Skill Security Checklist
• OpenClaw Skill Troubleshooting: 15 Common Errors

Next steps

• Install workflow: How to Install OpenClaw Skills
• Runtime failures: How to Fix OpenClaw Skill Permission Denied (2026)
• Ongoing troubleshooting: OpenClaw Skill Troubleshooting: 15 Common Errors

References

• npm CLI npx Docs
• Conventional Commits
• OWASP Secure Coding Practices