agentleFS
Alexander-TyagunovAlexander-Tyagunov/magician/skills/inscribe/SKILL.md

inscribe

Skill14 starsChanged 49 days agoMITView on GitHub

Turns a workflow into a reusable project skill (.claude/skills/NAME/SKILL.md). Use to scaffold a new SKILL.md.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Create a new project skill from a recurring pattern.

    line 11
  2. Distinctive:

    Prefer a bundled CLI or an existing tool.

    line 79
  3. Distinctive:

    Check preconditions inside the tool (the order exists, is paid, is not refunded yet) and return a clear error.

    line 92

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/inscribe — Write New Skills

Create a new project skill from a recurring pattern. Claude Code loads project skills from .claude/skills/<name>/SKILL.md in the repository, so that is where the new skill goes.

When to Use

  • User explicitly requests a new skill
  • You observe a workflow worth capturing and the user wants it saved as a skill

Process

  1. Understand the pattern — ask: "What is the behavior you want to capture as a skill? Describe it in a sentence or two." End your turn. Wait for their description before naming or drafting anything.
  2. Name it — one word, verb-like, lowercase (e.g., migrate, localize, document). Glob .claude/skills/*/SKILL.md first so the name doesn't collide with an existing project skill.
  3. Define the scope — what does the skill do, where does it start, where does it end
  4. Draft the skill:
---
# Required: a specific description naming the trigger context (used for auto-invocation).
# Use a >- block if the description contains ": " (colon + space).
description: <one sentence; name when/why to use this skill and the trigger context>
# Optional: short slug name (defaults to the directory name)
name: <name>
# Recommended: scope tools to reduce permission prompts, e.g. Read, Edit(./src/**), Bash(git status)
# Never grant bare Bash, Write, WebFetch or WebSearch, a whole CLI (Bash(git *)), or an
# interpreter or shell; scope each command, e.g. Bash(npm test *).
# allowed-tools: <comma-separated tools>
# For side-effectful or standalone skills, prevent silent auto-invocation:
# disable-model-invocation: true
# Optional: hint shown for arguments. Always a double-quoted string (write an inner " as \").
# argument-hint: "[arg description]"
# For heavy read-only skills, run in a forked context to keep the main thread lean:
# context: fork
---

# /<name> — <Title>

<One paragraph describing the skill's purpose and when to use it.>

## Process

1. <step>
2. <step>
...

## Obstacles

State how this skill surfaces obstacles. If it does dispatchable work, include the producer block. If it dispatches/consumes workers, include roll-up + detect-pattern + memorize (`ctx learn`) wiring. If it is terminal/utility, a one-line note that it reports to the human in prose. The block format is described in magician's lore/obstacles.md.

## Completion Signal

"<name> complete. <what was accomplished>."

Quote any other frontmatter value that contains : or #, or that starts with [ { | > ! & * ' or ", so it parses as a plain string.

  1. Write to disk — write the drafted content to .claude/skills/<name>/SKILL.md (writing the file creates the folders).
  2. Test it — invoke the skill once to verify it works as expected
  3. Commit — only after the user confirms they want it committed:
git add .claude/skills/<name>/
git commit -m "feat: add /<name> skill"
  1. Say: "Skill /<name> created in .claude/skills/<name>/. If /<name> does not show up yet, restart Claude Code."

If the skill needs an MCP tool

Prefer a bundled CLI or an existing tool. Each MCP tool definition costs context on every turn, and tool choice degrades above about 30–50 tools (tool budget). When the user still wants an MCP server, draft its tools to this shape:

{
  "name": "orders_refund",
  "description": "Refund one order. Use when the user asks to refund or cancel a paid order. Not for partial refunds.",
  "inputSchema": {"type": "object", "properties": {"order_id": {"type": "string"}}, "required": ["order_id"]},
  "annotations": {"readOnlyHint": false, "destructiveHint": true, "idempotentHint": true}
}
  • One tool per user intent, not one per API endpoint. Group the lookups behind one read tool and each change behind its own write tool.
  • Set readOnlyHint: true on every tool that changes nothing, so a client can allow it without a prompt. Set destructiveHint on a tool that deletes or overwrites.
  • Check preconditions inside the tool (the order exists, is paid, is not refunded yet) and return a clear error. A precondition written only in the skill is advisory.
  • Suggest deferred loading (ToolSearch) when the server is needed only now and then.

Skill Quality Checklist

  • Frontmatter uses only valid fields (description, name, allowed-tools, disable-model-invocation, argument-hint, context), parses as strict YAML, and has a specific description
  • allowed-tools is scoped: no bare Bash/Write/WebFetch/WebSearch, no whole-CLI or interpreter grants, file writes as Edit(<path>)
  • SKILL.md stays lean (<~250 lines); heavy reference material lives in references/ and is linked
  • Has a clear completion signal
  • Has an ## Obstacles section — a producer block if it does dispatchable work, roll-up/detect/memorize wiring if it consumes workers, or a one-line terminal note otherwise (see lore/obstacles.md)
  • Process steps are concrete and actionable (not vague)
  • Does not duplicate an existing skill
  • Name is memorable and matches the behavior
  • A rule that must hold every time is a hook, not only skill text: a skill is advisory (advisory vs enforced)
  • Any MCP tool has one intent, a readOnlyHint, and its preconditions checked in the tool

Obstacles

/inscribe is an interactive, user-facing scaffolder — it reports to the human, not upward — so it emits no upward Obstacles block: surface any blocker in prose to the user and stop. See lore/obstacles.md.

Completion Signal

"Inscribe complete. New skill /<name> ready. Invoke it with /<name>."

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.