fix(v2 audit): remove topic/ leftovers, add CONTRIBUTING.md

Systematic audit of v2 port found leftover v1 references and one broken link:

1. .claude/memory/MEMORY.md template — rewritten to remove "Topic Files"
   section and pointers to .claude/memory/topics/ (replaced by knowledge/ in v2).
   New "Knowledge Wiki" section documents the 6 subdirs and growth pattern.
   Architecture header changed: "Index + topic files" → "Hot cache + knowledge wiki".

2. CLAUDE.md line 205 (Context Save Protocol) — "create topic file" →
   "write a knowledge article in .claude/memory/knowledge/concepts/{slug}.md"

3. CLAUDE.md line 405 (Memory Entry Quality) — "needs more? → topic file" →
   "needs more? → knowledge article in knowledge/concepts/{slug}.md"

4. CONTRIBUTING.md — created. README linked to CONTRIBUTING.md which did not
   exist (broken link on first click). New file has minimal welcome + ground
   rules (zero deps, pure Markdown, Obsidian optional).

Known remaining warning (acceptable):
- lint.py flags 2 stale_experiments warnings on fresh clone for demo
  experiments 001 and 002. These are deleted during first-run setup per
  CLAUDE.md Step 5, so warnings disappear after setup.

Verified via grep that zero "topic file" / "topics/" strings remain outside
example content. README metrics unchanged: 7/7 P0 still passing.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: awrshift

.claude/memory/MEMORY.md

@@ -1,6 +1,6 @@
 # Project Memory
 Last updated: [DATE] (Session: #1)
-Architecture: Index (< 200 lines, auto-loaded) + topic files (on-demand)
+Architecture: Hot cache (< 200 lines, auto-loaded) + knowledge wiki (on-demand)
 
 ---
 
@@ -30,20 +30,23 @@ Architecture: Index (< 200 lines, auto-loaded) + topic files (on-demand)
 |----------|-----------|--------|
 | [Example: "Used raw SQL"] | [SQL injection in user input] | [Always use parameterized queries] |
 
-## Topic Files (read on-demand)
+## Knowledge Wiki (read on-demand)
 
-> As your project grows, move detailed knowledge into topic files.
-> MEMORY.md stays under 200 lines. Topics have no size limit.
-> Claude reads topic files when working on that area — not every session.
+> Deep knowledge lives in `.claude/memory/knowledge/`, not here.
+> MEMORY.md stays under 200 lines (Anthropic auto-load cap).
+> The wiki is read on-demand through `knowledge/index.md` — Claude reads individual articles only when relevant, not every session.
 
-| Topic | File | When to read |
-|-------|------|-------------|
-| [Example: API patterns] | `.claude/memory/topics/api.md` | Working on API endpoints |
-| [Example: Database] | `.claude/memory/topics/database.md` | Schema changes, queries |
-| [Example: Deployment] | `.claude/memory/topics/deployment.md` | CI/CD, hosting, env config |
+**Wiki entry point:** `.claude/memory/knowledge/index.md`
 
-> **How to create a topic file:**
-> 1. When a section above grows beyond 5-10 entries on the same theme
-> 2. Create `.claude/memory/topics/{topic-name}.md`
-> 3. Move detailed entries there, keep one-line summary here
-> 4. Add to the table above
+**Subdirectories:**
+- `concepts/` — single-topic deep dives
+- `connections/` — cross-concept relationships
+- `meetings/` — structured meeting index articles
+- `qa/` — filed query answers (compounding loop via `query.py --file-back`)
+- `projects/` — per-project wiki articles
+- `experiments/` — gravestone articles (stay after raw files archived)
+
+> **How to grow the wiki:**
+> 1. When a theme in MEMORY.md grows beyond 5-10 entries on the same topic, move details to `.claude/memory/knowledge/concepts/{slug}.md`
+> 2. Keep a one-line summary here with a pointer: `- **Theme name** [YYYY-MM] — see `knowledge/concepts/{slug}`
+> 3. Or let the session-end hook + `compile.py` do it automatically from `daily/` logs

CLAUDE.md

@@ -202,7 +202,7 @@ Triggers: user says "save context" / "update context", session is ending, or hoo
 
 Execute ALL 3 steps in order:
 
-1. **MEMORY.md** — add new verified patterns (with `[YYYY-MM]` date). Keep < 200 lines. If a theme grows > 5 entries → create topic file.
+1. **MEMORY.md** — add new verified patterns (with `[YYYY-MM]` date). Keep < 200 lines. If a theme grows > 5 entries → write a knowledge article in `.claude/memory/knowledge/concepts/{slug}.md`.
 2. **next-session-prompt.md** — update ONLY your `<!-- PROJECT:name -->` section: what was done, key decisions, `### IMMEDIATE NEXT` (exact first steps for next session).
 3. **JOURNAL.md** — update task statuses if any active tasks.
 
@@ -402,7 +402,7 @@ Before writing to MEMORY.md, self-check every entry:
 2. **Specific** — "always use parameterized queries" > "be careful with SQL"
 3. **Actionable** — changes your future behavior, not just states a fact
 4. **Not duplicate** — scan existing entries, update if exists
-5. **One line** — needs more? → topic file
+5. **One line** — needs more? → knowledge article in `knowledge/concepts/{slug}.md`
 
 ### MEMORY.md — What does NOT go here
 

CONTRIBUTING.md

@@ -0,0 +1,28 @@
+# Contributing to Claude Memory Kit
+
+Thank you for your interest in contributing. This project is an OSS starter kit for Claude Code, focused on persistent memory and structured context management.
+
+## Ways to contribute
+
+- **Report a bug** — open an issue with steps to reproduce
+- **Propose a feature** — open an issue describing the use case before writing code
+- **Improve docs** — PRs for README, CLAUDE.md, or the `/tour` skill are welcome
+- **Share a pattern** — open a discussion if you've found a memory or rules pattern that works well
+
+## Pull requests
+
+- Keep PRs focused. One feature or fix per PR.
+- Match the existing style (pure Markdown, zero dependencies, stdlib-only Python).
+- Test scripts locally before pushing: `python3 .claude/memory/scripts/lint.py` on your changes.
+- If you change behavior, update `CHANGELOG.md` and mention the change in the PR description.
+
+## Ground rules
+
+- **Zero dependencies.** Scripts use Python stdlib only. No `pip install`. No external services beyond `claude -p` subprocess.
+- **Pure Markdown for content.** Keep the wiki and memory plain `.md` so any editor works.
+- **Obsidian remains optional.** Don't add features that require Obsidian to be installed.
+- **Be kind in issues and PRs.** Assume good intent.
+
+## License
+
+By contributing, you agree that your contributions will be licensed under the MIT License (see [LICENSE](LICENSE)).