Skill creator enhanced

Skill Creator Enhanced

A skill for creating new skills and iteratively improving them.

Core Principles (read first)

Four rules that govern every skill produced by this kit. Bake them into the draft, do not treat them as optional polish.

1. Skills over prompts

Repeat tasks live in skills, not in one-off prompts. Before drafting, ask: "Has the user done this two or more times in recent sessions?" If yes, it belongs in a skill. When auditing user history, surface candidates with this framing:

Based on recent sessions, what tasks are being done repeatedly that should be skills instead of one-off prompts? For each, suggest the skill name and the context it needs.

2. Three layers, not two

A real skill has three layers. Most authors stop at layer 2, that is where leverage is left on the table.

Layer	What it is	Example
1. Description	When/why to trigger	YAML frontmatter description
2. Instructions	What to do, step by step	SKILL.md body
3. Tooling	Scripts, MCPs, file refs, verification commands the model should actually invoke	scripts/verify.py, references/schema.md, MCP tool names

When drafting any skill, explicitly answer: "What tool or script could turn a judgment step into a deterministic one?" If a step is checkable, write a script, do not ask the model to eyeball it. Pair generation with verification: skills that produce an output should also verify it (lint, grep, test, type-check).

Which layer does a change belong in? When you're fixing or improving a skill, route the change to the right layer instead of dumping everything in the body:

• A procedure — multi-step workflow, new data/file access, a templated output, anything reusable across runs → layer 2 (body) or layer 3 (a script). If it's deterministic, prefer a script.
• A mindset — reasoning approach, tone, judgment, a tool-use heuristic, "think about X before Y" → this is guidance, not procedure; keep it lean in the body.
• A triggering fix — the skill fired when it shouldn't, or didn't fire when it should → layer 1 (the description), not the body. The body is only read after triggering, so trigger fixes in the body are invisible.

3. Composable, not monolithic

Skills are reusable util functions, not god-objects. Each skill: one specific job, clear purpose, small enough to chain with others. When auditing existing skills, apply:

• Visibility (both keys are real Claude Code SKILL.md frontmatter, verified against docs.claude.com/docs/en/skills):
  • High-risk side effects (deploy, commit, send messages, external API writes): add disable-model-invocation: true. User can still trigger via /skill-name, but Claude cannot auto-fire and the description is not loaded into context (so it cannot influence Claude's decisions either).
  • Pure background knowledge users would never /run themselves: add user-invocable: false. Hides from the / menu; Claude can still invoke. Description stays in context.
• Deterministic vs non-deterministic: any step where the model is "interpreting" a fixed, repeatable operation should be moved into a script inside the skill folder. Code = same result every time, no token cost. Keep the model for steps that genuinely need judgment.
• De-duplication: if two skills share logic, extract the shared part into a callable script or a smaller composable skill that both invoke.

When proposing rewrites, output a short changelog: what changed and why.

To run the rule-3 audit deterministically:

python3 ${CLAUDE_PLUGIN_ROOT}/skills/skill-creator-enhanced/scripts/audit_skills.py [PATH ...]
# default scans ~/.claude/skills and ~/.claude/plugins
# flags: missing frontmatter, oversize descriptions, duplicate names,
#        no layer-3 tooling, risky verbs without disable-model-invocation
# exit 0 = clean, exit 1 = findings

4. Compounding self-update loop

A skill's first version is never its best. After every use, ask:

Is this a one-time fix, or should this be in the skill forever?

If the answer is "forever," update the skill in the same turn, do not let the lesson live only in chat. The loop: use skill, inspect output, update skill, repeat. Treat user corrections during a skill run as direct edits to SKILL.md, not as ephemeral feedback. When a session ends with a fix the model had to be told, that fix is a missing instruction.

When the user finishes a session that used a skill, proactively offer:

Want me to fold today's corrections back into the skill so this is handled automatically next time?

Keep a feedback ledger so a lesson is learned once, not re-learned. A recurring failure mode is re-proposing a change that was already tried and discarded. The ledger (scripts/ledger.py) is an append-only record of every attempt and its outcome, modeled on EvoSkill's feedback history. Use it during the improve loop:

# before proposing a change to a skill, read what was already tried:
python3 ${CLAUDE_PLUGIN_ROOT}/skills/skill-creator-enhanced/scripts/ledger.py \
  list --skill <name>
# anything marked `discarded` failed before — only re-propose if your idea is
# genuinely different, and say how it differs.

# after evaluating a change, record the outcome (improved/kept/discarded):
python3 .../ledger.py add --skill <name> --kind proposal \
  --text "what you changed" --outcome discarded --score-delta -0.05

# when a change improves the skill, archive prior open items for it — old
# failures are noise against the new baseline (this is the reset-on-improvement
# rule: keep the ledger focused on what's still unsolved):
python3 .../ledger.py reset --skill <name>

To surface raw correction candidates from prior transcripts (heuristic; the ledger is the curated version):

python3 ${CLAUDE_PLUGIN_ROOT}/skills/skill-creator-enhanced/scripts/extract_corrections.py \
  [--days N] [--project SUBSTR] [--json]
# scans ~/.claude/projects/**/*.jsonl for user messages matching
# correction patterns (no/don't/stop/instead/actually/wrong/revert/undo...)

Plugin hooks (hooks/hooks.json) automate this loop, but only in sessions where this skill is actually invoked. Invoking the skill (via /skill-name or auto-trigger) sets a per-session sentinel; the other hooks check it and no-op otherwise, so the plugin never acts on unrelated work:

• PreToolUse (matcher Skill, hook_skill_invoked.py) — when skill-creator-enhanced is invoked, marks the session active and surfaces any unreviewed pending corrections from the ledger as a nudge to fold them in.
• PostToolUse (matcher Edit|Write|MultiEdit, hook_validate.py) — in an active session only, whenever a SKILL.md is edited, its frontmatter is validated immediately; a broken edit is reported back so it can be fixed in the same turn (rule 2 enforced by the harness).
• SessionEnd (hook_corrections.py) — for an active session only, the transcript is scanned for corrections, appended to the ledger as pending entries, then the sentinel is cleared.

---

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. 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. 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: 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)
• 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. (This very SKILL.md does that — the eval, description-optimization, and platform details live in references/ and are linked below.)
• Reference files clearly from SKILL.md with guidance on when to read them
• For large reference files (>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

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).

---

Evaluate and iterate

Once you have a draft and test prompts, run the eval → grade → review → improve loop. The full procedure — spawning with-skill and baseline runs, capturing timing, grading, aggregating the benchmark, launching the viewer, reading feedback, and how to think about improvements across iterations — lives in references/evaluation.md. Read it before running test cases; it's a continuous sequence you shouldn't stop partway through. It also covers the optional blind A/B comparison.

The short version of the loop, for orientation:

• Draft or edit the skill
• Run claude-with-access-to-the-skill on the test prompts (with-skill vs baseline)
• Evaluate outputs with the user: eval-viewer/generate_review.py for qualitative review, plus the quantitative benchmark
• Improve the skill from feedback, rerun, repeat until you and the user are satisfied
• Expand the test set and try again at larger scale

Optimize the description

The description frontmatter is the primary mechanism that decides whether Claude invokes a skill, so tuning it pays off on every future invocation. After the skill is in good shape, offer to optimize it. The full procedure — generating realistic trigger eval queries, reviewing them with the user, running the background optimization loop (scripts/run_loop.py), and applying the winning description — lives in references/description-optimization.md, along with an explanation of how skill triggering actually works.

Package and present

When the skill is done, package it into a distributable .skill file:

python3 -m scripts.package_skill <path/to/skill-folder>

This validates the skill first and only packages if it passes. If the present_files tool is available, present the resulting .skill file to the user and point them at its path so they can install it. If present_files isn't available, just tell them where the file is.

Platform notes

The core workflow is the same everywhere, but Claude.ai (no subagents, often no browser) and Cowork (subagents but no display) change the mechanics — and updating an existing skill in a read-only location needs care. When running outside standard Claude Code, read references/platforms.md for the adaptations.

---

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. Read each when you reach the corresponding stage:

• references/evaluation.md — Running test cases, grading, benchmarking, the viewer, and the improvement loop (read before evaluating)
• references/description-optimization.md — Tuning the description for triggering accuracy (read before optimizing)
• references/platforms.md — Claude.ai / Cowork adaptations and updating an existing skill
• references/schemas.md — JSON structures for evals.json, grading.json, etc.

---

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!