skill-improvement
PassAudited by ClawScan on May 10, 2026.
Overview
This instruction-only skill is coherent for improving Claude skills, but users should review file edits and be cautious when verification runs subagents or target-skill tools.
This skill appears safe to install as an instruction-only skill. Use it on a copy of important skills when possible, review every proposed diff before applying changes, and be careful with verification tests that execute a target skill's scripts or share sensitive skill content with subagents.
Findings (3)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
If used carelessly, selected fixes could alter how an installed skill behaves later.
The skill can change other skill files, which can affect future agent behavior, but the workflow explicitly requires user selection and includes backups and diffs.
User chooses which issues to fix... Apply selected fixes to skill files... Create `.backup` before changes... Display diff for each modification
Review the diagnostic report and diffs before accepting edits, and keep the generated backups until the optimized skill is verified.
Testing an untrusted or unsafe target skill could run its tools or scripts during verification.
Verification may involve running scripts or tools from the skill being tested; this is purpose-aligned for testing, but the user should ensure the target skill is trusted or sandboxed.
Execute the [skill-name] skill on this task... Follow the skill's workflow. Use any scripts or tools as instructed.
Run execution tests only for trusted skills, or use a sandbox/test copy and require explicit approval before executing target-skill scripts.
Skill contents and test scenarios may be passed to background subagents during verification.
The verification process shares skill details and test prompts with subagents. This is disclosed and scoped to testing, but users should notice the inter-agent data flow.
Dispatch subagents (parallel)... task(subagent_type="explore", prompt="[trigger test]", run_in_background=true)
Avoid including secrets in skill files or test prompts, and review what information is sent to subagents when testing sensitive skills.