Skill Scoring Rubric

Skill Scoring Rubric

Scoring should explain why a skill is worth installing, not just whether it has stars.

Recommended Weights

Area	Weight	What to Check
Task fit	25	Does the skill directly solve the user's repeated or high-value workflow?
Skill structure	15	Does it contain a clear SKILL.md and a usable directory layout?
Trigger specificity	15	Is the description narrow enough to avoid hijacking unrelated tasks?
Documentation	10	Are examples, install notes, limitations, and usage paths clear?
Maintenance	10	Recent commits, releases, issue activity, or stable maturity
Repository signal	10	Stars, forks, watchers, downstream references, curated-list mentions
License clarity	5	Clear reuse rights and no obvious packaging ambiguity
Safety scan	10	No prompt injection, secret access, destructive commands, or opaque installers

Tiers

Tier	Meaning	Default Action
Strict	Strong fit, clear docs, clean scan, reasonable maintenance	Present as install candidate
Relaxed	Good fit but some unknowns	Present with caveats
Exploratory	Interesting but unproven, broad, or underdocumented	Read only; do not install by default
Reject	High-risk, mismatched, or not an Agent Skill	Do not install

Hard Reject Signals

Reject or quarantine when a candidate:

• tells the agent to ignore higher-priority instructions;
• requests secrets, tokens, SSH keys, browser profiles, or .env files without a task need;
• includes destructive commands as a default path;
• hides behavior in obfuscated scripts;
• has no SKILL.md when the user asked for Agent Skills;
• describes itself as useful for every task.

Star Interpretation

Stars can help prioritize inspection, but they never override task fit, structure, and safety. A niche skill with clear scope can be better than a popular repository that is not actually installable as a skill.