Artifact

Artifact Register

Published artifacts are hosted web pages whose only handle is a URL. Unregistered, that URL exists solely in the transcript of the session that made it — which is why an artifact produced three weeks ago is effectively lost even though the page is still live.

This skill keeps a durable, in-repo record of every artifact produced alongside the project's CPM work: what it is, why it was made, and what it is associated with. That record is what makes an artifact findable later, so it can be picked up, reviewed, or amended rather than rebuilt.

The register is a plain Markdown file — docs/artifacts/index.md — so it is greppable, diffable, reviewable, and travels with the repo.

This skill is stateless: it creates no progress file. Every action is a single bounded read-and-write.

Input

Parse $ARGUMENTS:

• No arguments — run the review flow: show the register, and offer to add, amend, or retire entries.
• Contains a URL — run the register flow for that URL. Any text alongside it is taken as the description and association hint (e.g. /cpm:artifact https://… auth flow explorer for the client review, from spec 12).
• Begins with list — print the register and stop. Read-only.
• Anything else — treat it as a search term and show matching entries (matched against name, why, and association), so /cpm:artifact auth finds the auth-related artifacts.

Process

Step 1: Read the register

Read docs/artifacts/index.md if it exists. If it does not, this is the first registration — create the directory and seed the file with the header from Register format below. Do not create the file merely to display an empty register; report "No artifacts registered yet" instead.

Step 2: Establish the four facts

Every entry records four things. Never invent any of them — ask.

1. URL — the artifact's address. Required; without it there is no entry to make.
2. Name — what it is, in a few words. Default to the artifact's own title when it can be read; confirm rather than assume.
3. Why — one sentence on why it was made and who it was for. This is the field that makes an entry worth having: a URL and a date do not tell you, six weeks later, whether a page is worth reopening.
4. Associated with — the CPM document(s) it belongs to, as repo-relative paths. Use — when an artifact genuinely stands alone; do not force an association that is not real.

When registering, offer to read the artifact so the name and a description can be proposed rather than typed. Confirm what is proposed before writing it — a page's own title is often not how its author would describe it.

For the association, Glob the CPM artifact directories and offer the plausible candidates rather than asking for a path to be typed:

• docs/specifications/[0-9]*-spec-*.md, docs/briefs/[0-9]*-brief-*.md, docs/plans/[0-9]*-plan-*.md
• docs/architecture/[0-9]*-adr-*.md, docs/epics/[0-9]*-epic-*.md
• docs/reviews/[0-9]*-review-*.md, docs/retros/[0-9]*-retro-*.md
• docs/communications/[0-9]*-*.md, docs/audits/[0-9]*-audit-*.md

Multi-select — an artifact drawing on several epics names all of them.

Step 3: Write the entry

Append a row to docs/artifacts/index.md, newest first (immediately under the table header). Use the Edit tool. Never rewrite rows other than the one being changed.

Re-registering a URL already in the register updates that row in place rather than adding a second one — the register holds one row per artifact, keyed by URL.

Step 4: Write the backlink

For each associated document, add or extend an **Artifacts**: field in its top-level metadata block, so the relationship is discoverable from either end:

**Artifacts**: [Auth flow explorer](https://claude.ai/code/artifact/…)

Where the field already exists, append to it. This is the reverse of the register's "Associated with" column: the register answers "what have we produced?", the backlink answers "what came out of this spec?". Both directions matter, and each is cheap.

If a named document cannot be found, say so and still write the register row — a missing backlink is a smaller loss than a lost URL.

Step 5: Report

Show the entry as written and the paths touched. Where the register was created for the first time, say so — it is a new tracked file the user will want to commit.

Register format

docs/artifacts/index.md:

# Artifact Register

Published artifacts produced alongside this project's CPM work. Newest first.
Maintained by `/cpm:artifact` — see the associated documents for the work each one came from.

| Artifact | URL | Registered | Associated with | Why |
|---|---|---|---|---|
| Auth flow explorer | https://claude.ai/code/artifact/… | 2026-07-25 | `docs/specifications/12-spec-auth.md` | Interactive walkthrough of the token-refresh path, built for the client review |

• Registered is the date the entry was made, not necessarily the date the artifact was published — say so if they differ and it matters.
• Associated with holds repo-relative paths in backticks, comma-separated, or — for a standalone artifact.
• Why is one sentence. Where an artifact needs more explanation than that, link the document that carries it rather than growing the cell.

Retiring an entry

An artifact that has been superseded or deleted is struck through, not removed — ~~Auth flow explorer~~ with the reason appended to Why. A register that silently drops entries cannot answer "what happened to that page?", which is one of the questions it exists to answer.

Guidelines

• The register is written when the artifact is made, not later. Anything relying on a memory of the URL has already failed — that is the problem this skill exists to solve. Skills that publish artifacts register them as part of publishing.
• Four facts or it is not worth writing. A bare URL is what the transcript already had. Name, why, and association are the entry's value.
• One row per artifact, keyed by URL. Re-registering updates in place; the register never accumulates duplicates of the same page.
• Never invent an association. A standalone artifact records —. A forced link to a loosely-related spec is worse than none, because it sends a future reader to the wrong document.
• Both directions, always. The register lists what was produced; the backlink in each associated document shows what that work produced. Neither substitutes for the other.
• The register is a record, not a lifecycle. This skill does not publish, update, or delete artifacts themselves — it records where they are and what they were for.