Claude Project Memory — Development skill for Claude Code

project-memory

A [Claude Code](https://claude.com/claude-code) skill for keeping a project's memory correct across sessions and models. It packages one durable documentation system into two skills: **`project-memory`** (the machinery — bootstrap, PRD, bins, templates, scripts and hooks) and **`docs-sync`** (the write arm — bring every doc back in line with what the code and the session actually did).

This is the memory / documentation half of a two-repo generalized skill set. The non-memory skills — security gates, model-quality tooling, and review / advisory skills — live in the sibling repo **[claude-skill-suite](https://github.com/Evan-Daruwalla/claude-skill-suite)**.

The system

Four artifacts, each with one job:

Artifact	Role
HANDOFF.md	The single always-current snapshot. A fresh session reads it first.
docs/.md	An append-only chronological build log — the ground truth. Point-in-time snapshots live inside it as dated entries.
PRD_ROADMAP.md	A standing plan a model executes one small task at a time.
.claude/project-memory/	Binned project memory so a session loads only the facts a task needs. important.md holds the critical facts and is injected by a hook; every other kind of fact (decisions, people, timeline, operations, experiments, hardware, code...) has its own bin.

The design goal: a session started cold — by any model, including a cheaper one — can read `HANDOFF.md`, pick up the plan, and execute without the owner filling gaps from memory.

The workflows, and which skill owns them

Record entry, handoff and drift-check moved from `project-memory` to `docs-sync` on 2026-09-09: they are the ones that EDIT docs, they are what people actually ask for ("update the docs"), and splitting them keeps each skill small enough to load one section instead of the whole file.

`project-memory` — the machinery:

• Bootstrap — stand the system up in a new project, seeded from real git h