Skip to main content
Status: Stable Version: 1.0.0 Author: Rifteo Tags: workflow
Installation 
rifteo-skills add skill-benchmark

​
Summary

Read any SKILL.md, score it across five weighted quality dimensions (100 pts total), run a static compatibility analysis against 50+ AI agent profiles, and get a ranked fix list before publishing.
  • Scores Trigger quality (25 pts), Instruction clarity (25 pts), Agent-agnostic design (20 pts), Self-containment (15 pts), and Output definition (15 pts). A score of 70+ means publishable; below 50 means silent failures on most agents
  • Maps the skill to 7 agent profile types (covering 54 named agents) and assigns Compatible / Partial / Likely broken for each
  • Returns a ranked fix list with one specific, actionable improvement per dimension
  • Offers to auto-fix the top issue and re-score immediately, or export the report as skill-benchmark-report.md

​
SKILL.md file

​
Skill Benchmark

Score a SKILL.md file across quality and compatibility dimensions, then return a ranked fix list for improvement before publishing.

​
When to Use This Skill

Use this skill when the user:
  • Asks to benchmark, score, or audit a skill (“is my skill any good?”)
  • Wants to check cross-agent compatibility (“will this work on Cursor / Windsurf / Gemini?”)
  • Needs a ranked fix list before publishing a SKILL.md to the community repo
  • Is comparing two skills or evaluating quality before a pull request submission
  • Wants to audit a SKILL.md for compatibility before opening a PR

​
What Does It Check?

The skill performs purely static analysis, no agents are installed or called. It reads the target SKILL.md, applies the scoring rubric from references/rubric.md, checks the skill against agent compatibility signals from references/agent-profiles.md, and produces the report using references/report-template.md.In scope:
  • Any SKILL.md in the local project or at a specified path
  • All 5 scoring dimensions (trigger quality, instruction clarity, agent-agnostic design, self-containment, output definition)
  • Compatibility with 54 named AI agents across 7 profile types
  • Auto-fix of the top-ranked issue on request
Out of scope:
  • Runtime testing of the skill against a live agent this is static analysis only
  • Scoring SKILL.md files for frameworks other than the Rifteo skills format

​
How It Works

Step 0: Locate the Target SkillSearch the current directory, then .claude/skills/, .agents/skills/, .cline/skills/, and global skill paths. If multiple files are found, list them and ask the user to confirm.Step 1: IntakeExtract the skill name, description, body line count, presence of bundled resources (scripts/, references/), and any explicit agent target mentions.Step 2: Score Each DimensionApply the rubric from references/rubric.md. For each of the 5 dimensions, record the score, a one-sentence rationale, and the single most impactful fix available.Step 3: Run Compatibility MatrixCheck the skill against each of the 7 agent profiles for hard-fail and soft-fail signals. Assign the worst result across all applicable profiles per named agent (FAIL overrides WARN overrides PASS).Step 4: Generate the ReportFill in the full report template: total score, per-dimension breakdown, compatibility matrix (54 agents grouped into Compatible / Partial / Broken), and ranked fix list.Step 5: Offer Next StepsAfter presenting the report, offer three options: auto-fix the top-ranked issue and re-score; deep-dive into any dimension with full evidence; or export the report as skill-benchmark-report.md.

​
Output

ScoreMeaning
70–100Publishable, skill is ready for the community repo
50–69Needs work some agents will degrade or skip steps
< 50Not publishable will silently fail on most agents
Example output structure:
Skill: my-skill
Total score: 74/100

Dimension scores:
 Trigger quality:    20/25
 Instruction clarity:  22/25
 Agent-agnostic design: 14/20
 Self-containment:   10/15
 Output definition:   8/15

Compatibility:
 Compatible (38 agents): Claude Code, Cursor, ...
 Partial (12 agents):  Windsurf, ...
 Likely broken (4 agents): ...

Top fixes (ranked):
 1. [Output definition] Add an explicit output format section: ...
 2. [Self-containment] Reference is missing: scripts/cvss-scorer.py ...
 3. [Agent-agnostic] Remove IDE-specific path assumption ...

​
Known Limitations

  • Scoring is strict: 70 is the publishable threshold, not 60 or 65
  • Compatibility results reflect static signals; a skill rated Partial may still work on some agents depending on their current version
  • The auto-fix option applies only the single top-ranked fix per run re-run to address the next issue

find-skills

Discover and install Rifteo skills from the community ecosystem

finding-writer

Convert raw pentest notes into structured audit findings ready for reporting

compliance-gap-analyzer

Aggregate findings into a gap report across ISO 27001, NIST CSF, PCI-DSS, and OWASP