skill-creator-plus

name	skill-creator-plus
description	Create, test, evaluate, and iteratively improve Claude skills. Use when users say "create a skill", "make a skill for", "write a SKILL.md", "turn this into a skill", "run evals", "test my skill", "benchmark my skill", "optimize my skill description", "improve triggering", "blind comparison", "A/B test my skill", or want to package a skill for distribution. Also triggers on "skill-creator", editing an existing skill, or reviewing skill quality.
metadata	{"author":"Yaniv Golan","version":"0.4.2","license":"MIT"}

Skill Creator

A skill for creating new skills and iteratively improving them.

At a high level, the process of creating a skill goes like this:

• Decide what you want the skill to do and roughly how it should do it
• Write a draft of the skill
• Create a few test prompts and run claude-with-access-to-the-skill on them
• Help the user evaluate the results both qualitatively and quantitatively
  • While the runs happen in the background, draft some quantitative evals if there aren't any (if there are some, you can either use as is or modify if you feel something needs to change about them). Then explain them to the user (or if they already existed, explain the ones that already exist)
  • Use the eval-viewer/generate_review.py script to show the user the results for them to look at, and also let them look at the quantitative metrics
• Rewrite the skill based on feedback from the user's evaluation of the results (and also if there are any glaring flaws that become apparent from the quantitative benchmarks)
• Repeat until you're satisfied
• Expand the test set and try again at larger scale

Your job when using this skill is to figure out where the user is in this process and then jump in and help them progress through these stages. So for instance, maybe they're like "I want to make a skill for X". You can help narrow down what they mean, write a draft, write the test cases, figure out how they want to evaluate, run all the prompts, and repeat.

On the other hand, maybe they already have a draft of the skill. In this case you can go straight to the eval/iterate part of the loop.

Of course, you should always be flexible and if the user is like "I don't need to run a bunch of evaluations, just vibe with me", you can do that instead.

Then after the skill is done (but again, the order is flexible), you can also run the skill description improver, which we have a whole separate script for, to optimize the triggering of the skill.

Cool? Cool.

Communicating with the user

The skill creator is liable to be used by people across a wide range of familiarity with coding jargon. If you haven't heard (and how could you, it's only very recently that it started), there's a trend now where the power of Claude is inspiring plumbers to open up their terminals, parents and grandparents to google "how to install npm". On the other hand, the bulk of users are probably fairly computer-literate.

So please pay attention to context cues to understand how to phrase your communication! In the default case, just to give you some idea:

• "evaluation" and "benchmark" are borderline, but OK
• for "JSON" and "assertion" you want to see serious cues from the user that they know what those things are before using them without explaining them

It's OK to briefly explain terms if you're in doubt, and feel free to clarify terms with a short definition if you're unsure if the user will get it.

---

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., they say "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 the gaps, and should confirm before proceeding to the next step.

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. Which use case category does this fall into? (see references/official-guide-patterns.md for details)
   • Document & Asset Creation — consistent, high-quality output (docs, presentations, code, designs)
   • Workflow Automation — multi-step processes benefiting from consistent methodology
   • MCP Enhancement — workflow guidance layered on top of MCP tool access For a more granular taxonomy (9 types from Anthropic's internal experience), see the "Expanded Skill Type Taxonomy" section of the reference file — it covers Library & API Reference, Product Verification, Data Fetching & Analysis, Business Process & Team Automation, Code Scaffolding & Templates, Code Quality & Review, CI/CD & Deployment, Runbooks, and Infrastructure Operations. Knowing the type helps choose the right techniques.
5. Should we set up test cases to verify the skill works? 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 need them. Suggest the appropriate default based on the skill type, but let the user decide.

Define Success Criteria

Before writing anything, help the user articulate what "working" looks like. These are aspirational targets, not precise thresholds — but they keep the iteration loop focused.

• Quantitative: Does the skill trigger on ~90% of relevant queries? Does it complete the workflow in fewer tool calls than without? Are there zero failed API calls?
• Qualitative: Can a user get through the workflow without needing to redirect Claude? Are results consistent across sessions? Does a new user succeed on their first try?

See references/official-guide-patterns.md (Success Criteria section) for measurement approaches.

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:

Portable fields (work on every agentskills.io host — Claude, Gemini CLI, Cursor, OpenCode, etc.):

• name: Skill identifier (kebab-case only, no spaces or capitals, must match the folder name)
• description: The single most important field. It's how agents decide whether to load the skill. Write it as a trigger, not a summary: [What it does] + [When to use it] + [Key capabilities]. Must be under 1024 characters, no XML tags. Make it assertive — models tend to undertrigger. On portable hosts this is the only discovery text the model sees, so it must stand alone. See references/official-guide-patterns.md (Description Field Formula) for the full formula and examples.
• license (optional): License name or bundled license file reference.
• compatibility (optional, max 500 chars): Use when your skill has environment requirements (e.g. "Requires git, docker"; "Designed for Claude Code").
• metadata (optional): Arbitrary key-value pairs. Recommended: author, version.
• allowed-tools (optional, experimental): Space-separated pre-approved tool patterns.

Claude-specific extensions (supported; skills remain portable if you don't use them):

• when_to_use (Claude-only): A separate field Claude Code joins with description in its skill listing. Non-Claude hosts ignore this field entirely — never put load-bearing trigger info here. If you use it, keep description self-sufficient and use when_to_use purely to add extra phrasing for Claude's matcher. The combined description + when_to_use is capped at 1,536 chars in Claude's listing.

• allowed-tools / hooks / shell permission-prompt rule (Claude-specific): On Claude Code, any non-empty value in allowed-tools, hooks, or shell triggers a user permission prompt on invocation. To keep a skill silently auto-allowed on Claude, leave these empty and rely on session permissions. Other hosts vary.

• Other Claude-only fields (model, effort, agent, context: fork, paths, disable-model-invocation, user-invocable, argument-hint, Dynamic Context Injection, path variables): documented in references/official-guide-patterns.md (Advanced Skill Authoring Features). Non-Claude hosts silently ignore them.

• the rest of the skill :)

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
    └── assets/     - Files used in output (templates, icons, fonts)

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 (<500 lines 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 500 lines; if you're approaching this limit, add an additional layer of hierarchy along with clear pointers about where the model using the skill should go next to follow up.
• Reference files clearly from SKILL.md with guidance on when to read them
• For large reference files (>300 lines), include a table of contents

For advanced patterns (setup/config, persistent data, on-demand hooks, Dynamic Context Injection, path variables like ${CLAUDE_SKILL_DIR}, relative markdown links), see references/official-guide-patterns.md (Practical Lessons and Advanced Skill Authoring Features sections).

Domain organization: When a skill supports multiple domains/frameworks, organize by variant:

cloud-deploy/
├── SKILL.md (workflow + selection)
└── references/
    ├── aws.md
    ├── gcp.md
    └── azure.md

Claude reads only the relevant reference file.

Technical Rules & Structural Patterns

See references/official-guide-patterns.md for: hard technical rules (SKILL.md naming, folder naming, forbidden frontmatter patterns), five structural patterns (Sequential Workflow, Multi-MCP, Iterative Refinement, Context-Aware, Domain-Specific), and the problem-first vs tool-first design choice.

Skills must not contain malware, exploit code, or anything that would surprise the user if described. Don't create misleading skills or skills designed to facilitate unauthorized access.

Writing Patterns

Prefer using the imperative form in instructions. Be specific and actionable — instead of "Validate the data before proceeding", write specific steps with actual commands and common failure modes. Include error handling for common failures. Reference bundled resources clearly. Bundle scripts for critical validations — code is deterministic, language interpretation isn't.

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

Explain the why behind instructions instead of heavy-handed MUSTs. Make skills general, not narrow to specific examples. Draft, then review with fresh eyes.

Key principles (see references/official-guide-patterns.md, "Practical Lessons" for full details): don't state the obvious (Claude already knows a lot), build a Gotchas section (highest-signal content), avoid railroading Claude (preserve flexibility), and store scripts so Claude composes rather than reconstructs boilerplate.

Script vs. Instruct

When designing a skill's architecture, decide what goes into bundled scripts/ vs. what stays as SKILL.md instructions. Use this as a first-pass heuristic at draft time:

Script when the work is:

• Deterministic and repeatable (data transforms, format conversion, file I/O)
• Validatable by a fixed rule or exit code (schema checks, regex, required fields)
• API calls with specific auth/endpoint details
• The kind of boilerplate Claude would otherwise re-derive every run

Instruct when the work needs:

• Judgment (tone, what to include/exclude, how to frame results)
• Context-dependent decisions (which approach fits this user's situation)
• Flexible error recovery (interpreting unexpected results, deciding next steps)
• Workflow orchestration where the sequence may vary

This is a starting point, not the final answer. The strongest signal for what to script comes later, from observing convergence across eval runs (see "Look for repeated work across test cases" below) — if 2-3 independent runs all reinvent the same helper, that's empirical evidence the logic belongs in a script. Don't over-script upfront; let the convergence signal guide you. See Script vs. Instruct decision framework ("When to Script vs. When to Instruct") for the full framework and examples.

When you do bundle a script, design it for agent consumption — non-interactive, --help-documented, structured output (JSON/CSV), helpful errors, meaningful exit codes, idempotent by default. A script that works fine for a human can be unusable for an agent. See official-guide-patterns.md ("Designing Scripts for Agent Use") for the full conventions.

Test Cases

Pro Tip from the official guide: Iterate on a single task before expanding. The most effective skill creators iterate on a single challenging task until Claude succeeds, then extract the winning approach into a skill. This leverages in-context learning and provides faster signal than broad testing. Once you have a working foundation, expand to multiple test cases for coverage.

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.

Per the official guide, effective testing covers three areas:

1. Triggering tests — Does the skill load at the right times? (obvious tasks, paraphrased requests, and confirming it doesn't trigger on unrelated topics)
2. Functional tests — Does the skill produce correct outputs? (valid outputs, API calls succeed, error handling works, edge cases covered)
3. Performance comparison — Does the skill actually improve results vs. baseline? (fewer tool calls, fewer user corrections, lower token usage)

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).

Running and evaluating test cases

This section is one continuous 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 viewer 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 the eval_metadata.json files and evals/evals.json with the assertions once drafted. Also explain to the user what they'll see in the viewer — 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 4: Grade, aggregate, and launch the viewer

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 config directory (e.g., eval-1/with_skill/grading.json). The grading.json expectations array must use the fields text, passed, and evidence (not name/met/details or other variants) — the viewer depends on these exact field names. 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. If a check is also something a user of the finished skill would benefit from running themselves (e.g., a validate_X.py or smoke_test_X.sh), bundle it in the skill's scripts/ directory so the same code serves both the eval grader and end users.

2. 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 the viewer expects. Put each with_skill version before its baseline counterpart.

3. 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.

4. Launch the viewer with both qualitative outputs and quantitative data:

   nohup python <skill-creator-path>/eval-viewer/generate_review.py \
     <workspace>/iteration-N \
     --skill-name "my-skill" \
     --benchmark <workspace>/iteration-N/benchmark.json \
     > /dev/null 2>&1 &
   VIEWER_PID=$!
   

   For iteration 2+, also pass --previous-workspace <workspace>/iteration-<N-1>.

   Cowork / headless environments: If webbrowser.open() is not available or the environment has no display, use --static <output_path> to write a standalone HTML file instead of starting a server. When the user clicks "Submit All Reviews", the viewer will download feedback.json and display the raw JSON in a copyable textarea. The user can either move the downloaded file into the workspace or paste the JSON directly into the chat. Important: In static mode, you cannot read feedback.json from disk — see "Cowork-Specific Instructions" below for how to handle the feedback loop.

Note: please use generate_review.py to create the viewer; there's no need to write custom HTML.

5. Tell the user something like: "I've opened the results in your browser. There are two tabs — 'Outputs' lets you click through each test case and leave feedback, 'Benchmark' shows the quantitative comparison. When you're done, come back here and let me know."

What the user sees in the viewer

The "Outputs" tab shows one test case at a time:

• Prompt: the task that was given
• Output: the files the skill produced, rendered inline where possible
• Previous Output (iteration 2+): collapsed section showing last iteration's output
• Formal Grades (if grading was run): collapsed section showing assertion pass/fail
• Feedback: a textbox that auto-saves as they type
• Previous Feedback (iteration 2+): their comments from last time, shown below the textbox

The "Benchmark" tab shows the stats summary: pass rates, timing, and token usage for each configuration, with per-eval breakdowns and analyst observations.

Navigation is via prev/next buttons or arrow keys. When done, they click "Submit All Reviews" which saves all feedback to feedback.json.

Step 5: Read the feedback

When the user tells you they're done, read feedback.json:

{
  "reviews": [
    {"run_id": "eval-0-with_skill", "feedback": "the chart is missing axis labels", "timestamp": "..."},
    {"run_id": "eval-1-with_skill", "feedback": "", "timestamp": "..."},
    {"run_id": "eval-2-with_skill", "feedback": "perfect, love this", "timestamp": "..."}
  ],
  "status": "complete"
}

Empty feedback means the user thought it was fine. Focus your improvements on the test cases where the user had specific complaints.

Kill the viewer server when you're done with it:

kill $VIEWER_PID 2>/dev/null

---

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. The big picture thing that's happening here is that we're trying to create skills that can be used a million times (maybe literally, maybe even more who knows) across many different prompts. Here you and the user are iterating on only a few examples over and over again because it helps move faster. The user knows these examples in and out and it's quick for them to assess new outputs. But if the skill you and the user are codeveloping 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, you might try branching out and using different metaphors, or recommending 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, you can try getting rid of the parts of the skill that are making it do that and seeing what happens.

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 and really make things happen. 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 what they actually wrote, and then transmit this understanding into the instructions. If you find yourself writing ALWAYS or NEVER in all caps, or using super rigid structures, that's a yellow flag — if possible, reframe and explain the reasoning so that the model understands why the thing you're asking for is important. That's a more humane, powerful, and effective approach.

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 to something. 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 saves every future invocation from reinventing the wheel. This works in both directions: scripts you bundle for end users (validators, smoke tests) can also be reused as eval-time grader assertions in later iterations. See Script vs. Instruct decision framework ("When to Script vs. When to Instruct") for guidance on what belongs in a script vs. what should stay as instructions.

5. Check against the official troubleshooting patterns. Consult references/official-guide-patterns.md (Troubleshooting Guide section) for common issues: instructions not followed (too verbose? buried? ambiguous?), skill not triggering (description too generic?), skill over-triggering (needs negative triggers or scope clarification?), large context degradation (SKILL.md too big? move content to references/).

This task is pretty important (we are trying to create billions a year in economic value here!) and your thinking time is not the blocker; take your time and really mull things over. I'd suggest writing a draft revision and then looking at it anew and making improvements. 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. Launch the reviewer with --previous-workspace pointing at the 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

Remember: Anthropic's own experience is that most of their best skills began as just a few lines and a single gotcha, then got better over time as Claude hit new edge cases. It's fine to ship something small and iterate — perfection on day one is not the goal.

---

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 for review using the HTML template:

1. Read the template from assets/eval_review.html
2. Replace the placeholders:
   • __EVAL_DATA_PLACEHOLDER__ → the JSON array of eval items (no quotes around it — it's a JS variable assignment). After serializing, replace </ with <\/ in the JSON string to prevent </script> in data from breaking the HTML parser.
   • __SKILL_NAME_PLACEHOLDER__ → the skill's name
   • __SKILL_DESCRIPTION_PLACEHOLDER__ → the skill's current description
3. Write to a temp file (e.g., /tmp/eval_review_<skill-name>.html) and open it: open /tmp/eval_review_<skill-name>.html
4. The user can edit queries, toggle should-trigger, add/remove entries, then click "Export Eval Set"
5. The file downloads to ~/Downloads/eval_set.json — check the Downloads folder for the most recent version in case there are multiple (e.g., eval_set (1).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 \
  --target-length 500 \
  --plateau-patience 2 \
  --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.

--target-length is a soft target, not a hard cap. The improver is told about it; selection breaks ties by preferring shorter descriptions. Default 500 chars — shrink it if the user has many skills installed and is at risk of Claude's listing-budget collapse; raise it to 800+ if trigger accuracy genuinely needs more surface area. The hard cap stays at 1,024 chars (agentskills.io description field limit).

--plateau-patience stops the loop early if the test score hasn't improved in N consecutive iterations, instead of burning all 5 iterations appending verbiage to chase edge cases. Default 2.

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

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 (or until plateau). 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 (with shorter winning ties) 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.

---

Validate Against the Official Checklist

Before packaging, run through the quick checklist from references/official-guide-patterns.md to catch common issues:

• Folder named in kebab-case
• SKILL.md file exists (exact spelling, case-sensitive)
• YAML frontmatter has --- delimiters
• name field: kebab-case, no spaces, no capitals, matches folder name
• description includes WHAT the skill does and WHEN to use it
• No XML tags (< >) anywhere in frontmatter
• No "claude" or "anthropic" in the skill name
• Instructions are clear and actionable (not vague)
• Error handling included for likely failure modes
• Examples provided where helpful
• References clearly linked from SKILL.md
• SKILL.md stays under ~500 lines (detailed content in references/)
• No README.md inside the skill folder

You can run python -m scripts.quick_validate <path-to-skill> to check some of these automatically.

Package and Present (only if present_files tool is available)

Check whether you have access to the present_files tool. If you don't, skip this step. If you do, package the skill and present the .skill file to the user:

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

After packaging, direct the user to the resulting .skill file path so they can install it.

---

Claude.ai-specific instructions

In Claude.ai, the core workflow is the same (draft → test → review → improve → repeat), but because Claude.ai doesn't have subagents, some mechanics change. Here's what to adapt:

Running test cases: No subagents means no parallel execution. For each test case, read the skill's SKILL.md, then follow its instructions to accomplish the test prompt yourself. Do them one at a time. This is less rigorous than independent subagents (you wrote the skill and you're also running it, so you have full context), but it's a useful sanity check — and the human review step compensates. Skip the baseline runs — just use the skill to complete the task as requested.

Reviewing results: If you can't open a browser (e.g., Claude.ai's VM has no display, or you're on a remote server), skip the browser reviewer entirely. Instead, present results directly in the conversation. For each test case, show the prompt and the output. If the output is a file the user needs to see (like a .docx or .xlsx), save it to the filesystem and tell them where it is so they can download and inspect it. Ask for feedback inline: "How does this look? Anything you'd change?"

Benchmarking: Skip the quantitative benchmarking — it relies on baseline comparisons which aren't meaningful without subagents. Focus on qualitative feedback from the user.

The iteration loop: Same as before — improve the skill, rerun the test cases, ask for feedback — just without the browser reviewer in the middle. You can still organize results into iteration directories on the filesystem if you have one.

Description optimization: This section requires the claude CLI tool (specifically claude -p) which is only available in Claude Code. Skip it if you're on Claude.ai.

Blind comparison: Requires subagents. Skip it.

Packaging: The package_skill.py script works anywhere with Python and a filesystem. On Claude.ai, you can run it and the user can download the resulting .skill file.

Updating an existing skill: The user might be asking you to update an existing skill, not create a new one. In this case:

• Preserve the original name. Note the skill's directory name and name frontmatter field -- use them unchanged. E.g., if the installed skill is research-helper, output research-helper.skill (not research-helper-v2).
• Copy to a writeable location before editing. The installed skill path may be read-only. Copy to /tmp/skill-name/, edit there, and package from the copy.
• If packaging manually, stage in /tmp/ first, then copy to the output directory -- direct writes may fail due to permissions.

---

Cowork-Specific Instructions

If you're in Cowork, the main things to know are:

• You have subagents, so the main workflow (spawn test cases in parallel, run baselines, grade, etc.) all works. (However, if you run into severe problems with timeouts, it's OK to run the test prompts in series rather than parallel.)
• You don't have a browser or display, so when generating the eval viewer, use --static <output_path> to write a standalone HTML file instead of starting a server. Then proffer a link that the user can click to open the HTML in their browser.
• Claude tends to skip the eval viewer in Cowork and jump straight to analyzing results itself. This defeats the purpose — the human needs to see the outputs and give feedback before you revise anything. Always run generate_review.py first (not your own custom HTML), then wait for the human to review. The eval viewer exists so the human can form their own opinion before you start making changes.
• Feedback loop workaround (IMPORTANT): In static mode there is no server, so the viewer cannot POST feedback to disk. When the user clicks "Submit All Reviews", the viewer downloads feedback.json to their browser and shows the raw JSON in a copyable textarea. You (Claude) cannot read browser downloads, so the feedback loop requires one of these:
  1. The user pastes the JSON directly into the chat — you parse it inline.
  2. The user moves feedback.json into the workspace folder you're using — you then read it with the Read tool. When you tell the user "come back and tell me you're done reviewing", also say: "The viewer will show your feedback as JSON — please paste it here or drop the downloaded feedback.json into [workspace path]." Don't assume the file will appear on its own.
• Packaging works — package_skill.py just needs Python and a filesystem.
• Description optimization (run_loop.py / run_eval.py) should work in Cowork just fine since it uses claude -p via subprocess, not a browser, but please save it until you've fully finished making the skill and the user agrees it's in good shape.
• Updating an existing skill: The user might be asking you to update an existing skill, not create a new one. Follow the update guidance in the claude.ai section above.

---

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/official-guide-patterns.md — Anthropic's official best practices: use case categories, description formula, five skill patterns, instructions best practices, technical rules, troubleshooting guide, and quick checklist. Consult this when designing a new skill or diagnosing issues with an existing one.

---

Repeating one more time the core loop here for emphasis:

• Figure out what the skill is about
• Draft or edit the skill
• Run claude-with-access-to-the-skill on test prompts
• With the user, evaluate the outputs:
  • Create benchmark.json and run eval-viewer/generate_review.py to help the user review them
  • Run quantitative evals
• Repeat until you and the user are satisfied
• Package the final skill and return it to the user.

Please add steps to your TodoList, if you have such a thing, to make sure you don't forget. If you're in Cowork, please specifically put "Create evals JSON and run eval-viewer/generate_review.py so human can review test cases" in your TodoList to make sure it happens.

Good luck!