docs: slim CLAUDE.md to a spine + topical pointer table

CLAUDE.md was 688L / 40KB and was injected into every session — most
of its bulk was topical pitfalls only relevant when working on a
specific subsystem (Prefab UI, typed cache, OpenAPI spec, API
helpers). Every session paid that context tax even for topics never
touched.

This commit completes the refactor started in the previous commit:

- Replace the ~293-line Known Pitfalls section with a short
  Cross-cutting list (5 genuinely repo-wide rules — null SKUs,
  generated files, raw list mocks, kwargs, worktree safety) plus a
  6-row Topical pointer table. Each row lists the most-greppable
  keywords inside (DataTable, archive, ListResponse, uv.lock,
  HEAD:refs/heads, …) so an agent grepping CLAUDE.md lands on the
  right pointer with enough context to follow the link.
- Delete the 65-line "API Response Handling Best Practices"
  section; the content lives in
  katana_public_api_client/docs/guide.md and is reachable via the
  topical table.
- Extend the Detailed Documentation table with rows for the three
  new subsystem docs added in the previous commit.

Final size: 381L / 22KB — a 45% reduction.

Also update stale section pointers in three harness files that
referenced sections that moved:

- .claude/agents/domain-advisor.md (4 citations now pointing at
  guide.md / spec-authoring.md / prefab/README.md).
- .claude/skills/techdebt/SKILL.md and
  .claude/skills/pre-commit/SKILL.md (anti-pattern source
  citations).

The "rebase before PR" rule that previously lived in CLAUDE.md is
already enforced by /open-pr and /review-pr CRITICAL sections, so
the topical table delegates to the skills rather than duplicating
the rule.

Co-Authored-By: Claude <noreply@anthropic.com>

Author: dougborg

.claude/agents/domain-advisor.md

@@ -53,7 +53,7 @@ items = unwrap_data(response, default=[])              # 200 list (.data envelop
 if is_success(response): ...                           # 201/204
 ```
 
-Anti-pattern: `if response.status_code == 200`. Source: `CLAUDE.md` API Response Handling Best Practices.
+Anti-pattern: `if response.status_code == 200`. Source: `katana_public_api_client/docs/guide.md` — Response Handling section.
 
 ### "How do I handle UNSET attrs fields?"
 
@@ -64,15 +64,15 @@ status = unwrap_unset(order.status, None)        # read attrs field; default if
 payload_value = to_unset(maybe_optional)         # write attrs field; UNSET if None
 ```
 
-Never `isinstance(value, type(UNSET))`, never `hasattr(order, 'status')` for attrs-defined fields. Source: `CLAUDE.md` Anti-Patterns.
+Never `isinstance(value, type(UNSET))`, never `hasattr(order, 'status')` for attrs-defined fields. Source: `katana_public_api_client/docs/guide.md` — Response Handling section (anti-patterns).
 
 ### "Where do retries go?"
 
 Transport layer (`katana_client.py`). All 100+ endpoints get retries, rate-limit handling, and pagination automatically. Never wrap individual API methods. Source: `katana_public_api_client/docs/adr/0001-transport-layer-resilience.md`.
 
 ### "How are list responses shaped?"
 
-Katana wraps every list in `{"data": [...]}`. Test mocks must reflect this — never define raw arrays. Source: `CLAUDE.md` Known Pitfalls.
+Katana wraps every list in `{"data": [...]}`. Test mocks must reflect this — never define raw arrays. Source: `katana_public_api_client/docs/spec-authoring.md` — "List responses must use a `ListResponse` schema".
 
 ### "Which exceptions can `unwrap_as` raise?"
 
@@ -84,11 +84,11 @@ Katana wraps every list in `{"data": [...]}`. Test mocks must reflect this — n
 | 5xx | `ServerError` |
 | Other 4xx | `APIError` |
 
-Source: `CLAUDE.md` Exception Hierarchy.
+Source: `katana_public_api_client/docs/guide.md` — Response Handling section, "Exception hierarchy".
 
 ### "When do I need to update the help resource?"
 
-When you add or change MCP tool parameters. The help resource at `katana_mcp_server/.../resources/help.py` has hardcoded tool documentation that drifts silently. Source: `CLAUDE.md` Known Pitfalls.
+When you add or change MCP tool parameters. The help resource at `katana_mcp_server/.../resources/help.py` has hardcoded tool documentation that drifts silently. Source: `katana_mcp_server/docs/prefab/README.md` — "Help resource drift".
 
 ## EDGE CASES
 

.claude/skills/pre-commit/SKILL.md

@@ -46,7 +46,7 @@ git diff --cached
 Look for:
 
 - Generated-file edits (`api/**/*.py`, `models/**/*.py`, `client.py`) without spec change
-- Anti-patterns from `CLAUDE.md` "Known Pitfalls" and "Anti-Patterns to Avoid"
+- Anti-patterns from `CLAUDE.md` "Known Pitfalls" and `katana_public_api_client/docs/guide.md` "Response Handling" section
 
 ### 3. Verify test coverage
 

.claude/skills/techdebt/SKILL.md

@@ -35,7 +35,7 @@ uv run poe quick-check
 Categories:
 
 1. **Dead code** — unused imports/vars/functions/classes; unreachable paths; commented-out blocks. Confirm with `LSP findReferences` before flagging.
-2. **Outdated patterns** — anti-patterns from `CLAUDE.md` Known Pitfalls / Anti-Patterns sections (UNSET misuse, manual status checks, retry wrapping, hasattr on attrs models, raw list mocks).
+2. **Outdated patterns** — anti-patterns from `katana_public_api_client/docs/guide.md` "Response Handling" section and `CLAUDE.md` "Known Pitfalls" (UNSET misuse, manual status checks, retry wrapping, hasattr on attrs models, raw list mocks).
 3. **Code duplication** — repeated logic that should be a helper, copy-pasted test setup that should be a fixture.
 4. **Code smells** — broad `except Exception`, missing type annotations on public functions, parameter explosions, name shadowing, circular imports.
 5. **Missing best practices** — public functions without docstrings, async tests missing `@pytest.mark.asyncio`, fixtures that should be `scope="session"`.