skill-creator-advance

Skill Creator Advance

A skill for creating new skills and iteratively improving them.

The core loop: draft → test → review → improve → repeat.

• Decide what the skill should do → write a draft
• Create test prompts → run claude-with-the-skill on them
• Evaluate results (qualitative review + quantitative assertions)
• Rewrite based on feedback → repeat until satisfied
• Optionally optimize the description for better triggering

Your job is to figure out where the user is in this process and help them progress. Maybe they want to start from scratch, or maybe they already have a draft. Be flexible — if they say "just vibe with me", skip the formal eval machinery.

Communicating with the user

Users range from non-technical to expert — plumbers opening terminals, grandparents googling "how to install npm", alongside experienced developers. Pay attention to context cues to gauge their familiarity:

• "evaluation" and "benchmark" are borderline, but generally OK
• For "JSON" and "assertion", look for serious cues the user knows these terms before using them without explanation

When in doubt, briefly define terms inline. It's better to over-explain once than to lose someone.

---

Creating a skill

Capture Intent

Start by understanding the user's intent. The current conversation might already contain a workflow the user wants to capture (e.g., "turn this into a skill"). If so, extract answers from the conversation history first — the tools used, the sequence of steps, corrections the user made, input/output formats observed. The user may need to fill gaps, and should confirm before proceeding.

Clarify these four questions:

1. What should this skill enable Claude to do?
2. When should this skill trigger? (what user phrases/contexts)
3. What's the expected output format?
4. Should we set up test cases? Skills with objectively verifiable outputs (file transforms, data extraction, code generation, fixed workflow steps) benefit from test cases. Skills with subjective outputs (writing style, art) often don't. Suggest the appropriate default, but let the user decide.

Interview and Research

Proactively ask questions about edge cases, input/output formats, example files, success criteria, and dependencies. Wait to write test prompts until you've got this part ironed out.

Check available MCPs - if useful for research (searching docs, finding similar skills, looking up best practices), research in parallel via subagents if available, otherwise inline. Come prepared with context to reduce burden on the user.

Write the SKILL.md

Based on the user interview, fill in these components:

• name: Skill identifier
• description: The primary triggering mechanism — include what the skill does AND when to use it. Claude tends to "undertrigger" skills, so make descriptions a little "pushy" (e.g., "Make sure to use this skill whenever the user mentions dashboards, data visualization, or internal metrics, even if they don't explicitly ask for a 'dashboard.'")
• compatibility: Required tools, dependencies (optional, rarely needed)
• the rest of the skill :)

Description Best Practices

Beyond being "pushy" to avoid under-triggering, good descriptions also include:

Negative triggers — Explicitly state when NOT to use the skill to avoid collisions with similar skills:

Do NOT use for domain-team skills (use skill-team instead). Do NOT use for simple file edits that don't need a skill.

Multilingual trigger keywords — If the skill's users speak multiple languages, add keywords at the end of the description in those languages. This helps Claude match the skill even when the user prompts in another language:

スキル作成・評価。技能建立・評估。

Before/after example:

• Before: "Create and test skills."
• After: "Create new skills, improve existing skills, and measure skill performance with iterative eval-driven development. Use when users want to create a skill from scratch, edit or optimize an existing skill, run evals, or benchmark performance. Do NOT use for domain-team convention skills (use skill-team). スキル作成・評価ループ。技能建立・評估迴圈。"

Skill Writing Guide

Anatomy of a Skill

skill-name/
├── SKILL.md (required)
│   ├── YAML frontmatter (name, description required)
│   └── Markdown instructions
└── Bundled Resources (optional)
    ├── scripts/    - Executable code for deterministic/repetitive tasks
    └── references/ - Docs loaded into context as needed

Each skill should also have a corresponding slash command entry point in the plugin's commands/ directory. See references/plugin-conventions.md for the format and examples.

Progressive Disclosure

Skills use a three-level loading system:

1. Metadata (name + description) - Always in context (~100 words)
2. SKILL.md body - In context whenever skill triggers (~6,000 tokens ideal)
3. Bundled resources - As needed (unlimited, scripts can execute without loading)

These word counts are approximate and you can feel free to go longer if needed.

Key patterns:

• Keep SKILL.md under ~6,000 tokens (~4,500 words); if approaching this limit, extract detailed content into reference files with clear pointers from SKILL.md.
• Reference files clearly from SKILL.md with guidance on when to read them
• For large reference files (>~8,000 tokens), include a table of contents

Domain organization: When a skill supports multiple domains/frameworks, organize by variant (e.g., references/aws.md, references/gcp.md) — Claude reads only the relevant reference file.

Working with Existing Plugin Ecosystems

When creating a skill that will live inside an existing plugin, match the plugin's conventions rather than imposing a new structure. Read references/plugin-conventions.md for detailed guidance on:

• Observing and matching the target plugin's style
• Choosing between lightweight, standard, and full directory structures
• Key conventions: relative paths, token budgets, bundled file organization

Principle of Lack of Surprise

This goes without saying, but skills must not contain malware, exploit code, or any content that could compromise system security. A skill's contents should not surprise the user in their intent if described. Don't go along with requests to create misleading skills or skills designed to facilitate unauthorized access, data exfiltration, or other malicious activities. Things like a "roleplay as an XYZ" are OK though.

Writing Patterns

Prefer using the imperative form in instructions.

Defining output formats - You can do it like this:

## Report structure
ALWAYS use this exact template:
# [Title]
## Executive summary
## Key findings
## Recommendations

Examples pattern - It's useful to include examples. You can format them like this (but if "Input" and "Output" are in the examples you might want to deviate a little):

## Commit message format
**Example 1:**
Input: Added user authentication with JWT tokens
Output: feat(auth): implement JWT-based authentication

Writing Style

Try to explain to the model why things are important in lieu of heavy-handed musty MUSTs. Use theory of mind and try to make the skill general and not super-narrow to specific examples. Start by writing a draft and then look at it with fresh eyes and improve it.

Test Cases

After writing the skill draft, come up with 2-3 realistic test prompts — the kind of thing a real user would actually say. Share them with the user: [you don't have to use this exact language] "Here are a few test cases I'd like to try. Do these look right, or do you want to add more?" Then run them.

Save test cases to evals/evals.json. Don't write assertions yet — just the prompts. You'll draft assertions in the next step while the runs are in progress.

{
  "skill_name": "example-skill",
  "evals": [
    {
      "id": 1,
      "prompt": "User's task prompt",
      "expected_output": "Description of expected result",
      "files": []
    }
  ]
}

See references/schemas.md for the full schema (including the assertions field, which you'll add later).

---

Improving an Existing Skill

When the user wants to improve a skill rather than build one from scratch, follow this flow instead of the full creation process:

1. Assess the Current State

Read the existing SKILL.md and all bundled files. Understand:

• What the skill does and how it's structured
• What conventions it follows (check the parent plugin's style)
• Known issues the user reports or you observe

2. Diagnose Improvement Areas

Look at these dimensions:

• Triggering: Is the description specific enough? Does it undertrigger or overtrigger?
• Instructions: Are they clear? Do they explain the "why"? Are there gaps?
• Structure: Is the directory organization appropriate for the skill's complexity?
• Coverage: Are edge cases handled? Are there missing workflows?
• Bundled files: Are reference files up to date? Are scripts working?

3. Propose Changes

Present a concise improvement plan to the user before making changes. Group changes by impact:

• High impact: Changes that affect the skill's core behavior or triggering
• Low impact: Cleanup, reorganization, wording improvements

4. Evaluate

Use the eval workflow (quick or full path, depending on complexity) to verify improvements. When improving an existing skill, the baseline should be the original version — snapshot it before editing (cp -r <skill-path> <workspace>/skill-snapshot/).

---

Running and evaluating test cases

Choosing Your Eval Path

Not every skill needs the full benchmark treatment. Choose based on complexity:

Quick eval path — For simple skills (formatters, templates, single-step workflows):

• Run 2-3 test cases manually (no subagent spawning)
• Review outputs inline with the user
• Skip grading, benchmarking, and baseline comparison
• Iterate based on direct user feedback
• Use this when the skill's quality is best judged by looking at it

Full eval path — For complex skills (multi-step workflows, skills with objective criteria):

• Spawn parallel subagent runs with baselines
• Grade with assertions, aggregate benchmarks
• Use structured inline review with markdown reports
• Use this when you need quantitative comparison or the skill has many moving parts

When in doubt, start with the quick path. You can always escalate to the full path if the quick path isn't giving enough signal.

---

The full eval path follows this sequence — don't stop partway through. Do NOT use /skill-test or any other testing skill.

Put results in <skill-name>-workspace/ as a sibling to the skill directory. Within the workspace, organize results by iteration (iteration-1/, iteration-2/, etc.) and within that, each test case gets a directory (eval-0/, eval-1/, etc.). Don't create all of this upfront — just create directories as you go.

Step 1: Spawn all runs (with-skill AND baseline) in the same turn

For each test case, spawn two subagents in the same turn — one with the skill, one without. This is important: don't spawn the with-skill runs first and then come back for baselines later. Launch everything at once so it all finishes around the same time.

With-skill run:

Execute this task:
- Skill path: <path-to-skill>
- Task: <eval prompt>
- Input files: <eval files if any, or "none">
- Save outputs to: <workspace>/iteration-<N>/eval-<ID>/with_skill/outputs/
- Outputs to save: <what the user cares about — e.g., "the .docx file", "the final CSV">

Baseline run (same prompt, but the baseline depends on context):

• Creating a new skill: no skill at all. Same prompt, no skill path, save to without_skill/outputs/.
• Improving an existing skill: the old version. Before editing, snapshot the skill (cp -r <skill-path> <workspace>/skill-snapshot/), then point the baseline subagent at the snapshot. Save to old_skill/outputs/.

Write an eval_metadata.json for each test case (assertions can be empty for now). Give each eval a descriptive name based on what it's testing — not just "eval-0". Use this name for the directory too. If this iteration uses new or modified eval prompts, create these files for each new eval directory — don't assume they carry over from previous iterations.

{
  "eval_id": 0,
  "eval_name": "descriptive-name-here",
  "prompt": "The user's task prompt",
  "assertions": []
}

Step 2: While runs are in progress, draft assertions

Don't just wait for the runs to finish — you can use this time productively. Draft quantitative assertions for each test case and explain them to the user. If assertions already exist in evals/evals.json, review them and explain what they check.

Good assertions are objectively verifiable and have descriptive names — they should read clearly in the benchmark report so someone glancing at the results immediately understands what each one checks. Subjective skills (writing style, design quality) are better evaluated qualitatively — don't force assertions onto things that need human judgment.

Update eval_metadata.json and evals/evals.json with assertions. Explain to the user what they'll see — both the qualitative outputs and the quantitative benchmark.

Step 3: As runs complete, capture timing data

When each subagent task completes, you receive a notification containing total_tokens and duration_ms. Save this data immediately to timing.json in the run directory:

{
  "total_tokens": 84852,
  "duration_ms": 23332,
  "total_duration_seconds": 23.3
}

This is the only opportunity to capture this data — it comes through the task notification and isn't persisted elsewhere. Process each notification as it arrives rather than trying to batch them.

Step 3.5: Self-assessment pass

Before grading and presenting results to the human, perform a quick automated check on each output. Read references/iteration-automation.md for the full protocol. In brief:

• Read each test case's output and check for obvious defects (empty output, format violations, crash artifacts)
• If a defect is clearly caused by a skill instruction issue, fix the skill and rerun that test case once
• Log results to self_assessment.json in each test case directory
• One pass only — no infinite repair loops

Step 4: Grade, aggregate, and present results

Once all runs are done:

1. Grade each run — spawn a grader subagent (or grade inline) that reads agents/grader.md and evaluates each assertion against the outputs. Save results to grading.json in each run directory. The grading.json expectations array must use the fields text, passed, and evidence (not name/met/details or other variants). For assertions that can be checked programmatically, write and run a script rather than eyeballing it — scripts are faster, more reliable, and can be reused across iterations.

2. Check for regressions (iteration 2+) — After grading, compare results against the previous iteration. Read references/iteration-automation.md for the full protocol. Lead with any regressions when reporting to the user.

3. Aggregate into benchmark — run the aggregation script from the skill-creator directory:

   python -m scripts.aggregate_benchmark <workspace>/iteration-N --skill-name <name>
   

   This produces benchmark.json and benchmark.md with pass_rate, time, and tokens for each configuration, with mean +/- stddev and the delta. If generating benchmark.json manually, see references/schemas.md for the exact schema. Put each with_skill version before its baseline counterpart.

4. Do an analyst pass — read the benchmark data and surface patterns the aggregate stats might hide. See agents/analyzer.md (the "Analyzing Benchmark Results" section) for what to look for — things like assertions that always pass regardless of skill (non-discriminating), high-variance evals (possibly flaky), and time/token tradeoffs.

5. Present results inline — For each test case, show the results directly in the conversation:

   ### Test Case: {eval_name}
   **Prompt:** {the task prompt}
   **Output:** {summary of key output files or inline content}
   **Grades:** {assertion pass/fail results with evidence, if graded}
   **Previous:** {what changed from last iteration, if iteration 2+}
   

   After presenting all test cases, show the benchmark summary (pass rates, timing, token usage).

6. Save a markdown report to <workspace>/iteration-N/review.md containing all test case results and benchmark data. This persists the results for cross-iteration comparison.

7. Ask for feedback — After presenting results, ask the user for feedback on each test case. Focus on the cases where they have specific complaints; no comment means it looked fine.

---

Improving the skill

This is the heart of the loop. You've run the test cases, the user has reviewed the results, and now you need to make the skill better based on their feedback.

How to think about improvements

1. Generalize from the feedback. You're trying to create skills that can be used a million times across many different prompts. You and the user are iterating on only a few examples because it helps move faster, but if the skill works only for those examples, it's useless. Rather than put in fiddly overfitty changes or oppressively constrictive MUSTs, if there's some stubborn issue, try branching out — use different metaphors, recommend different patterns of working. It's relatively cheap to try and maybe you'll land on something great.

2. Keep the prompt lean. Remove things that aren't pulling their weight. Make sure to read the transcripts, not just the final outputs — if it looks like the skill is making the model waste a bunch of time doing things that are unproductive, try getting rid of the parts of the skill that are causing that. The distinction matters: outputs tell you what happened, but transcripts tell you how it happened.

3. Explain the why. Try hard to explain the why behind everything you're asking the model to do. Today's LLMs are smart — they have good theory of mind and when given a good harness can go beyond rote instructions. Even if the feedback from the user is terse or frustrated, try to actually understand the task and why the user is writing what they wrote, and then transmit this understanding into the instructions. If you find yourself writing ALWAYS or NEVER in all caps, that's a yellow flag — reframe and explain the reasoning so that the model understands why it matters.

4. Look for repeated work across test cases. Read the transcripts from the test runs and notice if the subagents all independently wrote similar helper scripts or took the same multi-step approach. If all 3 test cases resulted in the subagent writing a create_docx.py or a build_chart.py, that's a strong signal the skill should bundle that script. Write it once, put it in scripts/, and tell the skill to use it.

This task is important and your thinking time is not the blocker; take your time and really mull things over. Write a draft revision and then look at it anew with fresh eyes. Really do your best to get into the head of the user and understand what they want and need.

The iteration loop

After improving the skill:

1. Apply your improvements to the skill
2. Rerun all test cases into a new iteration-<N+1>/ directory, including baseline runs. If you're creating a new skill, the baseline is always without_skill (no skill) — that stays the same across iterations. If you're improving an existing skill, use your judgment on what makes sense as the baseline: the original version the user came in with, or the previous iteration.
3. Present results inline and save to review.md, noting changes from previous iteration
4. Wait for the user to review and tell you they're done
5. Read the new feedback, improve again, repeat

Keep going until:

• The user says they're happy
• The feedback is all empty (everything looks good)
• You're not making meaningful progress

---

Advanced: Blind comparison

For situations where you want a more rigorous comparison between two versions of a skill (e.g., the user asks "is the new version actually better?"), there's a blind comparison system. Read agents/comparator.md and agents/analyzer.md for the details. The basic idea is: give two outputs to an independent agent without telling it which is which, and let it judge quality. Then analyze why the winner won.

This is optional, requires subagents, and most users won't need it. The human review loop is usually sufficient.

---

Description Optimization

The description field in SKILL.md frontmatter is the primary mechanism that determines whether Claude invokes a skill. After creating or improving a skill, offer to optimize the description for better triggering accuracy.

Step 1: Generate trigger eval queries

Create 20 eval queries — a mix of should-trigger and should-not-trigger. Save as JSON:

[
  {"query": "the user prompt", "should_trigger": true},
  {"query": "another prompt", "should_trigger": false}
]

The queries must be realistic and something a Claude Code or Claude.ai user would actually type. Not abstract requests, but requests that are concrete and specific and have a good amount of detail. For instance, file paths, personal context about the user's job or situation, column names and values, company names, URLs. A little bit of backstory. Some might be in lowercase or contain abbreviations or typos or casual speech. Use a mix of different lengths, and focus on edge cases rather than making them clear-cut (the user will get a chance to sign off on them).

Bad: "Format this data", "Extract text from PDF", "Create a chart"

Good: "ok so my boss just sent me this xlsx file (its in my downloads, called something like 'Q4 sales final FINAL v2.xlsx') and she wants me to add a column that shows the profit margin as a percentage. The revenue is in column C and costs are in column D i think"

For the should-trigger queries (8-10), think about coverage. You want different phrasings of the same intent — some formal, some casual. Include cases where the user doesn't explicitly name the skill or file type but clearly needs it. Throw in some uncommon use cases and cases where this skill competes with another but should win.

For the should-not-trigger queries (8-10), the most valuable ones are the near-misses — queries that share keywords or concepts with the skill but actually need something different. Think adjacent domains, ambiguous phrasing where a naive keyword match would trigger but shouldn't, and cases where the query touches on something the skill does but in a context where another tool is more appropriate.

The key thing to avoid: don't make should-not-trigger queries obviously irrelevant. "Write a fibonacci function" as a negative test for a PDF skill is too easy — it doesn't test anything. The negative cases should be genuinely tricky.

Step 2: Review with user

Present the eval set to the user inline for review. For each query, show:

| # | Query | Should Trigger? |
|---|-------|-----------------|
| 1 | "ok so my boss sent me this xlsx..." | ✅ Yes |
| 2 | "can you analyze the trends in..." | ❌ No |

Ask the user to confirm, edit, add, or remove queries before proceeding. Save the final eval set to <workspace>/trigger-eval.json.

This step matters — bad eval queries lead to bad descriptions.

Step 3: Run the optimization loop

Tell the user: "This will take some time — I'll run the optimization loop in the background and check on it periodically."

Save the eval set to the workspace, then run in the background:

python -m scripts.run_loop \
  --eval-set <path-to-trigger-eval.json> \
  --skill-path <path-to-skill> \
  --model <model-id-powering-this-session> \
  --max-iterations 5 \
  --verbose

Use the model ID from your system prompt (the one powering the current session) so the triggering test matches what the user actually experiences.

While it runs, periodically tail the output to give the user updates on which iteration it's on and what the scores look like.

This handles the full optimization loop automatically. It splits the eval set into 60% train and 40% held-out test, evaluates the current description (running each query 3 times to get a reliable trigger rate), then calls Claude to propose improvements based on what failed. It re-evaluates each new description on both train and test, iterating up to 5 times. When it's done, it opens an HTML report in the browser showing the results per iteration and returns JSON with best_description — selected by test score rather than train score to avoid overfitting.

How skill triggering works

Understanding the triggering mechanism helps design better eval queries. Skills appear in Claude's available_skills list with their name + description, and Claude decides whether to consult a skill based on that description. The important thing to know is that Claude only consults skills for tasks it can't easily handle on its own — simple, one-step queries like "read this PDF" may not trigger a skill even if the description matches perfectly, because Claude can handle them directly with basic tools. Complex, multi-step, or specialized queries reliably trigger skills when the description matches.

This means your eval queries should be substantive enough that Claude would actually benefit from consulting a skill. Simple queries like "read file X" are poor test cases — they won't trigger skills regardless of description quality.

Step 4: Apply the result

Take best_description from the JSON output and update the skill's SKILL.md frontmatter. Show the user before/after and report the scores.

---

Packaging

After the skill is complete, package it for distribution:

python -m scripts.package_skill <path/to/skill-folder>

This creates a .skill file that can be shared and installed.

---

Platform Adaptations

The instructions above assume Claude Code. If you're running in Claude.ai or Cowork, some mechanics change (no subagents, no browser, etc.). Read references/platform-adaptations.md for the platform-specific adjustments.

---

Reference files

The agents/ directory contains instructions for specialized subagents. Read them when you need to spawn the relevant subagent.

• agents/grader.md — How to evaluate assertions against outputs
• agents/comparator.md — How to do blind A/B comparison between two outputs
• agents/analyzer.md — How to analyze why one version beat another

The references/ directory has additional documentation:

• references/schemas.md — JSON structures for evals.json, grading.json, etc.
• references/plugin-conventions.md — Plugin ecosystem conventions, directory structures, and slash command format
• references/iteration-automation.md — Self-assessment and auto-regression detection protocols
• references/eval-methodology.md — Eval methodology principles: why the eval workflow is designed this way (optional)
• references/platform-adaptations.md — Claude.ai and Cowork platform-specific adjustments
• references/mermaid-usage-guidelines.md — When to use Mermaid diagrams vs prose in skill authoring (decision trees / state machines / routing), syntax conventions, cost-benefit framework

---

Core loop reminder: Draft → Test → Evaluate (inline review + evals) → Improve → Repeat → Package. Good luck!