memory-view

Memory View

A viewer for Claude Code's per-project auto-memory. This skill is a thin wrapper around a deterministic, zero-dependency TypeScript script: the script reads the memory vault, renders a single self-contained HTML explorer, and opens it in the browser. That is the whole job.

The model's only role is to run the script and open the report. It does not read, audit, edit, merge, split, delete, or otherwise curate memory. The HTML is a local, secret-redacted, human-facing artifact — same lineage and posture as session-report.

This is a user-invoked command (/memory-view); it is never auto-invoked. The optional project is passed as $project.

What it operates on

• Memory lives at ~/.claude/projects/<slug>/memory/, where <slug> is the project's git-root path with / and . both replaced by -. There is no global store — most projects have no memory dir, which is normal, not an error.
• MEMORY.md is the always-loaded index (budget: first 200 lines or 25 KB, whichever comes first). Topic files are markdown + YAML frontmatter (name, description, type ∈ user/feedback/project/reference), with bodies cross-linked via [[wikilinks]].

How to run

Zero-dependency TypeScript, no build step. Run from the skill's scripts/ dir. Detect the runtime (command -v node bun) — the user needs only one. Always pass --open so the report opens in the browser.

# DEFAULT — no project given: cross-project explorer, every memory grouped by project:
node   scan-memory.ts --open
bun run scan-memory.ts --open
npx tsx scan-memory.ts --open        # older Node without native type-stripping

# A specific project — single-project report:
node scan-memory.ts my-work --open   # bare name ⇒ single-project view
node scan-memory.ts --view --open    # current project (resolved from cwd git root)

The single stdout line is the HTML path; stats and warnings go to stderr.

Picking the project

• No project mentioned → run the default (--open, no positional): the all-projects explorer. This is the /memory-view case.
• User names a project (e.g. "show me memories for my-work") → pass it as a bare positional: node scan-memory.ts my-work --open. The script resolves it by matching the name against known project slugs (suffix match first, then substring), so a short name like my-work is enough. It also accepts an encoded slug, a project path, or a memory-dir path.
• User means "this project" / the current repo → use --view with no positional; the script resolves from the cwd's git root.
• If the resolved project has no memory (absent / empty / index-only), the script prints a short notice and lists the projects that do have memory — relay that so the user can pivot.