Agent skill

Refine

by romeerez in romeerez/orchid-orm

A skill your agent uses when the user prompts "refine design".

MITAuto-check passedDatabases

Install Refine

skills CLI
$ npx skills add romeerez/orchid-orm --skill refine -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install romeerez/orchid-orm refine --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/romeerez/orchid-orm.git skills-src && mkdir -p .claude/skills && cp -r skills-src/.agents/skills/refine .claude/skills/refine && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
refine
GitHub stars
543
Token cost
~2.2k tokens
SKILL.md length
1,206 words
Files
2
Skills in repo
13
Repo updated
First seen
Licence
MIT

At a glance

A skill your agent uses when the user prompts "refine design".

  • Works in 2 steps: Directly from the user → From another workflow
  • The user prompts refine design
  • Instructions only: no scripts, shell commands, URLs or credentials in SKILL.md
  • Tasks that involve ORMs and data access

What it does

Refine is an agent skill from romeerez/orchid-orm. Use when the user prompts "refine design".

Its SKILL.md is about 2.2k tokens, which your agent loads only when the skill is triggered. The skill folder holds 2 other files (for example `agents/openai.yaml`).

It sits in Databases, covering ORMs and data access. The licence is MIT.

When your agent uses it

  • The user prompts refine design
  • Tasks that involve ORMs and data access

Example prompts

  • “refine design”
  • “/refine”

Workflow steps

2 steps, taken from the first numbered list in SKILL.md.

  1. Directly from the user
  2. From another workflow

What it can do on your machine

Read from SKILL.md and the folder at commit f819fb5. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    No scripts in the folder and no shell commands in SKILL.md (its code samples are markdown).

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Refine loads about 2.2k tokens when it runs. Until then it costs about 12 tokens; SKILL.md has 1,206 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~12
When it runs · the whole SKILL.md, loaded when a task matches
~2.2k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from romeerez/orchid-orm at commit f819fb5, republished under its MIT licence (© romeerez). 1,206 words, ~2,231 tokens.

Download SKILL.mdSave it as .claude/skills/refine/SKILL.md (or your agent's skills folder). This skill also uses 1 other file; get the full folder from GitHub.
name
refine
description
Use when the user prompts "refine design".

Run an interactive refinement session for an existing feature design document.

This command is for clarifying and tightening a proposal that already exists. It is not a one-shot rewrite. The workflow may span multiple user turns.

The normal behavior is:

  • read the proposal carefully
  • identify what looks unclear, ambiguous, or under-specified
  • discuss questions with the user
  • record only user-confirmed questions in a ## Refinement section
  • record only user-confirmed conclusions under those questions
  • keep the main design above ## Refinement synchronized with every confirmed conclusion

Input

This workflow can start in two ways:

  1. Directly from the user

    The prompt argument should identify:

    • a feature or change folder
    • an idea number or idea title
    • optionally, a file name to refine

    If the user does not specify a file, default to:

    changes/<feature-name>/<NUMBER-idea-name>/selected-variant.md

    Examples:

    • /refine 611-row-level-security-integration 2
    • /refine row-level-security-integration "Run work inside an explicit RLS context"
    • /refine row-level-security-integration 2 variants.md
  2. From another workflow

    If the calling workflow already knows the input file, use that file directly instead of re-resolving it.

    By default, update that same file in place.

    Only write to a different output path when the caller or the user explicitly instructs you to do so.

Goal

Make the design document more reliable for later work by resolving important questions collaboratively and keeping both:

  • the main design
  • the refinement log

in sync throughout the session.

The refinement log is part of the artifact. It should help a later reader understand which questions mattered and what was decided.

Steps

  1. Resolve the input file

    If another workflow already provided the input file path, use it.

    If the user initiated the workflow directly:

    • find the best matching folder in changes/
    • resolve the idea folder inside changes/<feature-name>/
    • default the input file name to selected-variant.md unless the user specified another file

    Matching rules:

    • prefer an exact change-folder match
    • prefer an exact idea number match from folders such as 2-run-work-inside-an-explicit-rls-context
    • otherwise match the idea title against the folder suffix or ideas.md

    If multiple folders or ideas are plausible, ask one focused clarifying question. Do not guess.

    If the resolved input file does not exist, stop and tell the user. Do not create the starting design file in this command.

  2. Read the proposal and supporting context

    Read the full input file before commenting on it.

    Understand:

    • the goal
    • the proposed approach
    • the explicit assumptions
    • what is still implied rather than stated
    • which statements depend on existing orchid-orm behavior or external facts

    Read only the extra context that is needed to understand or verify the design:

    • relevant orchid-orm docs in docs/src/.vitepress/dist/llms.txt
    • external sources when needed for technical accuracy or when the user asks you to research

    When researching online, prefer official docs and trustworthy primary sources.

  3. Ensure the document has a refinement section

    If the input file does not already contain ## Refinement, append an empty ## Refinement section at the end of the file immediately.

    If ## Refinement already exists, preserve it and continue working inside it.

    There must be exactly one top-level ## Refinement section in the final document.

  4. Give the user your initial assessment

    After you understand the proposal, tell the user:

    • whether the design looks clear overall
    • which parts seem ambitious, risky, vague, or incomplete
    • whether anything important appears to be missing

    You may propose refinement questions at this point.

    Do not write those proposed questions into the document unless the user confirms that they should be tracked.

  5. Add only confirmed questions to ## Refinement

    When the user confirms a question to track, write it into ## Refinement unless that same question is already present.

    Use one subsection per tracked question in this form:

    md
    ## Refinement
    
    ### 1. <question title>
    
    <explain the question>
    
    Pending.

    Rules:

    • number questions in the order they are first accepted
    • do not duplicate a question that is already present
    • if the document already contains refinement questions, continue from the existing numbering
    • preserve earlier answered questions unless the user explicitly wants them removed
  6. Work through questions collaboratively

    Process questions one by one. Let the user steer the order when they want to.

    Different question types need different behavior:

    • Technical questions

      • inspect orchid-orm code, tests, docs, and external references as needed
      • prefer official documentation when online research is needed
      • distinguish confirmed facts from inference
    • Ergonomics or product-shape questions

      • propose options when helpful
      • include short code examples if they clarify the trade-offs
      • do not settle the decision on your own when it depends on user preference

    While a question is still open, you may temporarily write notes under it such as:

    • Pending.
    • candidate options
    • small example snippets
    • research notes that help the discussion

    Keep these notes concise and useful.

  7. Only record an answer after user confirmation

    Do not treat a tentative discussion as a final answer.

    Once the user confirms the conclusion for a question:

    • replace temporary notes or options under that question with a final answer
    • keep the final answer directly under the corresponding question

    The final recorded answer should include:

    • the conclusion
    • why it was chosen (if available)
    • the key points of how you reached it

    Use this form:

    md
    ### Question 1: <question>
    
    #### Answer:
    
    <final conclusion, why it was chosen, and the key evidence or reasoning behind it>
  8. Synchronize the main design after every confirmed answer

    After every confirmed answer, immediately update the document above ## Refinement so it matches the new conclusion.

    Update whichever parts are affected:

    • goal
    • approach
    • assumptions
    • examples
    • references
    • any other affected design details

    If a confirmed answer changes the design, the main document must be updated before you move on.

    Remove contradictions and stale statements instead of leaving the log to explain them away.

  9. Continue until the open questions are resolved

    The user may add more questions later in the same session.

    Each time:

    • add newly confirmed questions to ## Refinement
    • research or discuss them
    • wait for user confirmation
    • record the final answer
    • sync the design above ## Refinement

    If the user asks you to investigate something in the codebase or online, do that work before recording the answer.

  10. Finish cleanly

Show full SKILL.md (242 more words)Show less

When all tracked questions are answered, make sure the document is coherent as a whole.

The refinement log should still remain in the file unless the user explicitly asks to remove it.

Quality check

Before finishing, verify:

  • the correct input file was chosen
  • direct invocations defaulted to selected-variant.md unless the user chose another file
  • delegated invocations used the already-known input file unless explicitly told otherwise
  • the full proposal was read before assessment
  • ## Refinement exists exactly once
  • proposed questions were not written unless the user confirmed them
  • duplicate questions were not added
  • every recorded answer was user-confirmed
  • every recorded answer includes the conclusion and why or supporting basis when available
  • temporary options or notes were replaced once a final answer was recorded
  • the design above ## Refinement was updated after each confirmed conclusion
  • the main design and the refinement log do not contradict each other

Guardrails

  • Do not invent questions and silently write them into the log
  • Do not answer ergonomics decisions on the user's behalf
  • Do not leave the main design stale after recording a confirmed answer
  • Do not rely on memory for technical claims that should be checked in code, docs, or official external sources
  • Do not dump raw research notes into the document
  • Do not remove earlier refinement history unless the user asks
  • Do not create the initial proposal file here if it does not already exist
  • Ask a focused clarifying question whenever the target change, idea, or file is ambiguous

© romeerez, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 1 other file in .agents/skills/refine of romeerez/orchid-orm.

  • SKILL.md
  • agents/openai.yaml

Open the folder on GitHubat commit f819fb5

Compare with similar skills

Refine next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Refine compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Refine this skillromeerez/orchid-orm543—~2.2kAutomated safety check: PassMIT
Content Create Hero Imageprisma/web1.1k—~6.9kAutomated safety check: PassNone
Sea Orm 2FlyinPancake/yoink112—~2.9kAutomated safety check: PassApache-2.0
Prisma Client APIcurvenote/curvenote1702 repos~1.6kAutomated safety check: PassMIT
DB Migratesimstudioai/sim30k—~2kAutomated safety check: PassApache-2.0
DB Migrationskurealnum/dotfiles290—~820Automated safety check: PassNone

Similar skills

  • Official

    A skill your agent uses when the operator wants a hero or meta image for a Prisma blog post; asks to create or generate a blog hero, cover, social card, Open Graph, or YouTube image; mentions cover…

    1.1k GitHub stars~6.9k tokensUpdated today
    DatabasesAuto-check passed
  • Sea Orm 2

    FlyinPancake/yoink

    Expert guidance for SeaORM 2.0, Rust's async ORM with strongly-typed columns, nested ActiveModels, Entity Loader API, and entity-first workflow.

    112 GitHub stars~2.9k tokensUpdated 5 days ago
    DatabasesAuto-check passed
  • Prisma Client API

    curvenote/curvenote

    Prisma Client API reference covering model queries, filters, operators, and client methods.

    170 GitHub starsUsed in 2 repos~1.6k tokens
    DatabasesAuto-check passed
  • DB Migrate

    simstudioai/sim

    Author or review a Drizzle DB migration for zero-downtime safety — expand/contract phasing, backward-compatibility with the deployed app version, and writing the -- migration-safe acknowledgment the…

    30k GitHub stars~2k tokensUpdated today
    DatabasesAuto-check passed
  • DB Migrations

    kurealnum/dotfiles

    A skill your agent uses when generating or regenerating Drizzle migration files, changing database schema tables or columns, resolving migration sequence conflicts after rebase, reviewing migration…

    290 GitHub stars~820 tokensUpdated 5 mo ago
    DatabasesAuto-check passed
  • Change Database Schema

    martin-ueding/geo-activity-playground

    How to change the SQLAlchemy data model and generate the matching Alembic migration.

    100 GitHub stars~225 tokensUpdated 12 days ago
    DatabasesAuto-check passed

More from romeerez/orchid-orm

All 13 skills in this repo
  • Spec

    romeerez/orchid-orm

    A skill your agent uses when the user prompts "write spec" or "make spec".

    543 GitHub stars~2k tokensUpdated yesterday
    Auto-check passed
  • Task List

    romeerez/orchid-orm

    A skill your agent uses when user asks to write a task list, not to do a task

    543 GitHub stars~2.1k tokensUpdated yesterday
    Auto-check passed
  • Type Optimizer

    romeerez/orchid-orm

    A skill your agent uses when need to optimize TypeScript types.

    543 GitHub stars~798 tokensUpdated yesterday
    Auto-check passed
  • Code Doc

    romeerez/orchid-orm

    A skill your agent uses when the user prompts "code doc" to create or update internal Orchid ORM code documentation from changes/ specs, short-code feature folders, or existing implementation code.

    543 GitHub stars~2.2k tokensUpdated yesterday
    Auto-check passed
  • Ideas

    romeerez/orchid-orm

    A skill your agent uses when the user prompts "write ideas" or "make ideas".

    543 GitHub stars~2.4k tokensUpdated yesterday
    Auto-check passed
  • Implemenation Note

    romeerez/orchid-orm

    A skill your agent uses when the user prompts "implementation note" for an existing change idea.

    543 GitHub stars~1.3k tokensUpdated yesterday
    Auto-check passed

Categories

Questions about Refine

What does Refine do?

A skill your agent uses when the user prompts "refine design". Refine is an agent skill from romeerez/orchid-orm. Use when the user prompts "refine design".

When should I use Refine?

Refine fits situations like: the user prompts refine design; tasks that involve ORMs and data access.

How do I install Refine in Claude Code?

Run `npx skills add romeerez/orchid-orm --skill refine -a claude-code`. Or copy the skill folder (.agents/skills/refine in romeerez/orchid-orm) into .claude/skills/refine in your project. Claude Code loads it when a task matches its description.

How do I install Refine in Codex?

Run `npx skills add romeerez/orchid-orm --skill refine -a codex`. Or copy the skill folder (.agents/skills/refine in romeerez/orchid-orm) into .agents/skills/refine in your project. Codex loads it when a task matches its description.

Can I use Refine in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add romeerez/orchid-orm --skill refine -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/refine, .gemini/skills/refine, .github/skills/refine and .opencode/skills/refine in your project.

What does Refine need to run?

SKILL.md names no scripts, command-line tools or credentials: Refine is instructions for the agent only.

Does Refine access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Refine safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Refine use?

Refine is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Refine use?

About 2.2k tokens (SKILL.md is roughly 8.9k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Refine?

Skills that share tags, products or a category with Refine: Content Create Hero Image (prisma/web, 1.1k stars), Sea Orm 2 (FlyinPancake/yoink, 112 stars), Prisma Client API (curvenote/curvenote, 170 stars) and DB Migrate (simstudioai/sim, 30k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Refine?

romeerez (a GitHub user) maintains it in romeerez/orchid-orm, which has 543 GitHub stars. The repository holds 13 skills in this directory. The repository was last updated on October 8, 2026.

Source: romeerez/orchid-orm on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.