CtrlK
BlogDocsLog inGet started
Tessl Logo

mcollina/skill-optimizer

Optimizes AI skills for activation, clarity, and cross-model reliability. Use when creating or editing skill packs, diagnosing weak skill uptake, reducing regressions, tuning instruction salience, improving examples, shrinking context cost, or setting benchmark and release gates for skills. Trigger terms: skill optimization, activation gap, benchmark skill, with/without skill delta, regression, context budget, prompt salience.

87

1.14x
Quality

87%

Does it follow best practices?

Impact

87%

1.14x

Average score across 5 eval scenarios

SecuritybySnyk

Passed

No known issues

Overview
Quality
Evals
Security
Files

criteria.jsonevals/scenario-3/

{
  "context": "Tests whether the agent applies context-budget optimization correctly: keeping trigger lists, decision tables, and one integrated example while trimming repeated background explanations and duplicate examples; using the layered structure (SKILL.md as high-signal overview, rules/*.md for details); and applying safe simplification techniques.",
  "type": "weighted_checklist",
  "checklist": [
    {
      "name": "Trigger list preserved",
      "description": "SKILL-trimmed.md retains a trigger signal list (keywords that activate the skill, like REST endpoint, OpenAPI, API reference)",
      "max_score": 8
    },
    {
      "name": "Background prose trimmed",
      "description": "SKILL-trimmed.md does NOT contain the repeated 'Good API documentation is...' introduction paragraph or the 'Writing good docs takes practice' general advice section",
      "max_score": 8
    },
    {
      "name": "Duplicate examples removed",
      "description": "SKILL-trimmed.md contains at most ONE full endpoint example (the original had three with the same lesson), not all three repeated examples",
      "max_score": 8
    },
    {
      "name": "Single integrated example retained",
      "description": "SKILL-trimmed.md or one of the rules files contains exactly one complete endpoint example showing parameters table, response JSON, and errors table together",
      "max_score": 8
    },
    {
      "name": "Decision table or checklist kept",
      "description": "SKILL-trimmed.md retains the endpoint format checklist or parameters/errors table format (a structured reference, not prose description of it)",
      "max_score": 8
    },
    {
      "name": "Layered structure created",
      "description": "Both rules/endpoint-format.md AND rules/error-codes.md are created as separate files (not everything collapsed into SKILL-trimmed.md)",
      "max_score": 10
    },
    {
      "name": "SKILL.md links to rules",
      "description": "SKILL-trimmed.md contains links to rules/endpoint-format.md and rules/error-codes.md (not just mentions — actual markdown links)",
      "max_score": 10
    },
    {
      "name": "Paragraphs converted to checklists",
      "description": "At least one section that was prose paragraphs in the original (e.g. parameter description guidance) is converted to a bullet checklist in the trimmed version",
      "max_score": 8
    },
    {
      "name": "Rationale moved to rules",
      "description": "Explanatory rationale ('important because...', 'reduces cognitive load', 'why this matters') is NOT in SKILL-trimmed.md but may appear in rules files",
      "max_score": 8
    },
    {
      "name": "audit-notes what was cut",
      "description": "audit-notes.md explicitly names what was removed (e.g. 'removed duplicate examples', 'removed introductory background')",
      "max_score": 8
    },
    {
      "name": "audit-notes what was kept",
      "description": "audit-notes.md explicitly states what was preserved and why (e.g. 'kept trigger list because...', 'kept one integrated example because...')",
      "max_score": 8
    },
    {
      "name": "Token reduction achieved",
      "description": "SKILL-trimmed.md is shorter than the original SKILL-original.md (fewer lines or characters)",
      "max_score": 8
    }
  ]
}

evals

SKILL.md

tile.json