How do I install the Reference skill?

Run npx skillmds@latest add ajitta/reference in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Reference safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Reference?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Reference free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Reference?

ajitta (@ajitta) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Reference by ajitta — Agent Skill

Name: Reference
Author: ajitta

Reference

Point at a reference — treat a provided example as a map, not an answer, and prove comprehension before implementing. Use when the user says "use this as a reference", "make it like this", or supplies example code, a mockup, or a competitor's UX to build from.

ajitta 22aca2f 2.8 KB Updated Sep 17, 2026

File contents

Reference — A Reference Is Another Map

Best way to give a model a map: give it another map. Origin: Point at a Reference — see skills/loop/references/talk-source.md.

Iron Rule

Reference is not an answer to copy verbatim — it is material for understanding intent and behavior. Before implementing, prove comprehension with a semantics map.

Procedure

Read reference to the end; summarize intent, core behavior, invariants.
Present 4-category analysis:
- Behaviors to preserve — the reference's reason to exist
- Parts to transform for current environment — language/framework/scale differences
- Unnecessary or dangerous parts — must not be brought into this project
- Parts improvable beyond the reference — where we can do better
For port/transform work, also present proof of comprehension (semantics map): key reference excerpts ↔ new-environment counterparts side by side, gotcha points where behavior subtly differs, and an edge-case table with a Match column — identical / equivalent / changed ("equivalent" = same decision, different surface). Number every note and row so a correction can name one.
Contrast with current project conventions (code style, dependencies, test approach); present an application plan.
Sign-off gate — nothing is implemented until the user signs off. They reply semantics confirmed, or correct any row by its number ("note 5", "budget exhaustion row") and the map is revised before any code. In-session, offer the same choice once with AskUserQuestion: implement now / revise the map / stop.
On sign-off, port the reference's existing tests first, then implement. Record intentional deviations from the reference per /unknowns:notes (or /notes for copied installs).

Output

Artifact tool if available → else .unknowns/<YYYY-MM-DD>-reference-<slug>.html → else markdown; always echo the assembled reply in chat. Numbered mapping rows (reference excerpt ↔ counterpart plan) and edge-case rows each carry approve / request-change; selections assemble into a semantics confirmed reply or a numbered correction template. Details: skills/loop/references/output-routing.md

What Can Be a Reference

Existing implementation code, same algorithm in another language, working HTML mockup, competitor product UX, past project design docs, test code, screenshots/video, actual example of desired output. E.g., an HTML mockup as the map when building a React component.

---name: referencedescription: Point at a reference — treat a provided example as a map, not an answer, and prove comprehension before implementing. Use when the user says "use this as a reference", "make it like this", or supplies example code, a mockup, or a competitor's UX to build from.---# Reference — A Reference Is Another MapBest way to give a model a map: **give it another map**.Origin: Point at a Reference — see skills/loop/references/talk-source.md.## Iron Rule**Reference is not an answer to copy verbatim — it is material for understanding intent and behavior.**Before implementing, **prove comprehension** with a semantics map.## Procedure1. Read reference to the end; summarize **intent, core behavior, invariants**.2. Present 4-category analysis: - **Behaviors to preserve** — the reference's reason to exist - **Parts to transform for current environment** — language/framework/scale differences - **Unnecessary or dangerous parts** — must not be brought into this project - **Parts improvable beyond the reference** — where we can do better3. For port/transform work, also present **proof of comprehension (semantics map)**: key reference excerpts ↔ new-environment counterparts side by side, gotcha points where behavior subtly differs, and an edge-case table with a **Match** column — identical / equivalent / changed ("equivalent" = same decision, different surface). Number every note and row so a correction can name one.4. Contrast with current project conventions (code style, dependencies, test approach); present an **application plan**.5. **Sign-off gate — nothing is implemented until the user signs off.** They reply `semantics confirmed`, or correct any row by its number ("note 5", "budget exhaustion row") and the map is revised before any code. In-session, offer the same choice once with AskUserQuestion: implement now / revise the map / stop.6. On sign-off, **port the reference's existing tests first**, then implement. Record intentional deviations from the reference per `/unknowns:notes` (or `/notes` for copied installs).## OutputArtifact tool if available → else `.unknowns/<YYYY-MM-DD>-reference-<slug>.html` → elsemarkdown; always echo the assembled reply in chat.Numbered mapping rows (reference excerpt ↔ counterpart plan) and edge-case rows eachcarry **approve / request-change**; selections assemble into a `semantics confirmed`reply or a numbered correction template.Details: skills/loop/references/output-routing.md## What Can Be a ReferenceExisting implementation code, same algorithm in another language, working HTMLmockup, competitor product UX, past project design docs, test code,screenshots/video, actual example of desired output. E.g., an HTML mockup as themap when building a React component.

Frequently asked questions