Skill Creator

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
• Evaluate the results
• which can be through automated evals, but also it's totally fine and good for them to be evaluated by the human by hand and that's often the only way
• Rewrite the skill based on feedback from the evaluation
• 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.

Cool? Cool.

Building Blocks

The skill-creator operates on composable building blocks. Each has well-defined inputs and outputs.

| Building Block | Input | Output | Agent | |-----------|-------|--------|-------| | Eval Run | skill + eval prompt + files | transcript, outputs, metrics | agents/executor.md | | Grade Expectations | outputs + expectations | pass/fail per expectation | agents/grader.md | | Blind Compare | output A, output B, eval prompt | winner + reasoning | agents/comparator.md | | Post-hoc Analysis | winner + skills + transcripts | improvement suggestions | agents/analyzer.md |

Eval Run

Executes a skill on an eval prompt and produces measurable outputs.

• Input: Skill path, eval prompt, input files
• Output: transcript.md, outputs/, metrics.json
• Metrics captured: Tool calls, execution steps, output size, errors

Grade Expectations

Evaluates whether outputs meet defined expectations.

• Input: Expectations list, transcript, outputs directory
• Output: grading.json with pass/fail per expectation plus evidence
• Purpose: Objective measurement of skill performance

Blind Compare

Compares two outputs without knowing which skill produced them.

• Input: Output A path, Output B path, eval prompt, expectations (optional)
• Output: Winner (A/B/TIE), reasoning, quality scores
• Purpose: Unbiased comparison between skill versions

Post-hoc Analysis

After blind comparison, analyzes WHY the winner won.

• Input: Winner identity, both skills, both transcripts, comparison result
• Output: Winner strengths, loser weaknesses, improvement suggestions
• Purpose: Generate actionable improvements for next iteration

---

Environment Capabilities

Check whether you can spawn subagents — independent agents that execute tasks in parallel. If you can, you'll delegate work to executor, grader, comparator, and analyzer agents. If not, you'll do all work inline, sequentially.

This affects which modes are available and how they execute. The core workflows are the same — only the execution strategy changes.

---

Mode Workflows

Building blocks combine into higher-level workflows for each mode:

| Mode | Purpose | Workflow | |------|---------|----------| | Eval | Test skill performance | Executor → Grader → Results | | Improve | Iteratively optimize skill | Executor → Grader → Comparator → Analyzer → Apply | | Create | Interactive skill development | Interview → Research → Draft → Run → Refine | | Benchmark | Standardized performance measurement (requires subagents) | 3x runs per configuration → Aggregate → Analyze |

See references/mode-diagrams.md for detailed visual workflow diagrams.

---

Task Tracking

Use tasks to track progress on multi-step workflows.

Task Lifecycle

Each eval run becomes a task with stage progression:

pending → planning → implementing → reviewing → verifying → completed
          (prep)     (executor)     (grader)    (validate)

Creating Tasks

When running evals, create a task per eval run:

TaskCreate(
    subject="Eval 0, run 1 (with_skill)",
    description="Execute skill eval 0 with skill and grade expectations",
    activeForm="Preparing eval 0"
)

Updating Stages

Progress through stages as work completes:

TaskUpdate(task, status="planning")     # Prepare files, stage inputs
TaskUpdate(task, status="implementing") # Spawn executor subagent
TaskUpdate(task, status="reviewing")    # Spawn grader subagent
TaskUpdate(task, status="verifying")    # Validate outputs exist
TaskUpdate(task, status="completed")    # Done

Comparison Tasks

For blind comparisons (after all runs complete):

TaskCreate(
    subject="Compare skill-v1 vs skill-v2"
)
# planning = gather outputs
# implementing = spawn blind comparators
# reviewing = tally votes, handle ties
# verifying = if tied, run more comparisons or use efficiency
# completed = declare winner

---

Architecture

The coordinator (this skill):

1. Asks the user what they want to do and which skill to work on
2. Determines workspace location (ask if not obvious)
3. Creates workspace and tasks for tracking progress
4. Delegates work to subagents when available, otherwise executes inline
5. Tracks the best version (not necessarily the latest)
6. Reports results with evidence and metrics

Agent Types

| Agent | Role | Reference | |-------|------|-----------| | Executor | Run skill on a task, produce transcript + outputs + metrics | agents/executor.md | | Grader | Evaluate expectations against transcript and outputs | agents/grader.md | | Comparator | Blind A/B comparison between two outputs | agents/comparator.md | | Analyzer | Post-hoc analysis of comparison results | agents/analyzer.md |

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. 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.

Interview and Research

Proactively ask questions about edge cases, input/output formats, example files, success criteria, and dependencies.

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.

Initialize

Run the initialization script:

scripts/init_skill.py  --path 

This creates:

• SKILL.md template with frontmatter
• scripts/, references/, assets/ directories
• Example files to customize or delete

Fill SKILL.md Frontmatter

Based on interview, fill:

• name: Skill identifier
• description: When to trigger, what it does. This is the primary triggering mechanism - include both what the skill does AND specific contexts for when to use it. All "when to use" info goes here, not in the body. Note: currently Claude has a tendency to "undertrigger" skills -- to not use them when they'd be useful. To combat this, please make the skill descriptions a little bit "pushy". So for instance, instead of "How to build a simple fast dashboard to display internal Anthropic data.", you might write "How to build a simple fast dashboard to display internal Anthropic data. Make sure to use this skill whenever the user mentions dashboards, data visualization, internal metrics, or wants to display any kind of company data, even if they don't explicitly ask for a 'dashboard.'"
• compatibility: Required tools, dependencies (optional, rarely needed)

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)

What NOT to include: README.md, INSTALLATION_GUIDE.md, CHANGELOG.md, or any auxiliary documentation. Skills are for AI agents, not human onboarding.

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 (300 lines), include a table of contents

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.

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

Immediate Feedback Loop

Always have something cooking. Every time user adds an example or input:

1. Immediately start running it - don't wait for full specification
2. Show outputs in workspace - tell user: "The output is at X, take a look"
3. First runs in main agent loop - not subagent, so user sees the transcript
4. Seeing what Claude does helps user understand and refine requirements

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.

If the user wants evals, create evals/evals.json with this structure:

{
  "skill_name": "example-skill",
  "evals": [
    {
      "id": 1,
      "prompt": "User's task prompt",
      "expected_output": "Description of expected result",
      "files": [],
      "assertions": [
        "The output includes X",
        "The skill correctly handles Y"
      ]
    }
  ]
}

You can initialize with scripts/init_json.py evals evals/evals.json and validate with scripts/validate_json.py evals/evals.json. See references/schemas.md for the full schema.

Transition to Automated Iteration

Once gradable criteria are defined (expectations, success metrics), Claude can:

• More aggressively suggest improvements
• Run tests automatically (via subagents in the background if available, otherwise sequentially)
• Present results: "I tried X, it improved pass rate by Y%"

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:

scripts/package_skill.py 

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

---

Improving a skill

When user asks to improve a skill, ask:

1. Which skill? - Identify the skill to improve
2. How much time? - How long can Claude spend iterating?
3. What's the goal? - Target quality level, specific issues to fix, or general improvement

Claude should then autonomously iterate using the building blocks (run, grade, compare, analyze) to drive the skill toward the goal within the time budget.

Some advice on writing style when improving a skill:

1. Try to generalize from the feedback, rather than fixing specific examples one by one. 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 fiddley 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.

1. 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.

1. Last but not least, 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 and go beyond rote instructions and really make things

…

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: tim-hua-01
• Source: tim-hua-01/comment-on-docx
• License: MIT

Install and usage instructions live in the source repository linked above.