S1: Colony backend skeleton — Axum + SQLite, channels + messages CRUD

Monorepo structure:
- crates/colony-types: API types (serde + ts-rs), separate from DB models
- crates/colony: Axum server, SQLite via sqlx, migrations

Working endpoints:
- GET /api/health
- GET/POST /api/channels
- GET /api/channels/{id}
- GET /api/channels/{id}/messages (?since=, ?type=, ?user_id=)
- POST /api/channels/{id}/messages (with type + metadata)

Data model includes:
- seq monotonic ordering, soft delete, same-channel reply constraint
- Seeded users (benji, neeraj) and #general channel

Also: codex-review skill, .gitignore

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

.claude/skills/codex-review/SKILL.md

@@ -0,0 +1,36 @@
+---
+name: codex-review
+description: Get a parallel GPT-5.4 review of code, plans, or decisions. Use when you want a second opinion, finished a non-trivial implementation, or want to validate an approach before presenting to the ape.
+argument-hint: "[what to review — file paths, plan description, or question]"
+---
+
+# Codex Review
+
+Get an independent GPT-5.4 review via codex CLI. Use this proactively after finishing non-trivial work, or when you want a second opinion before proceeding.
+
+## When to use
+
+- After writing a significant chunk of code
+- Before presenting an architecture decision to the ape
+- When you're unsure about an approach
+- After a refactor to catch regressions
+- When the critic skill feels too heavy (critic is for decisions, this is for code review)
+
+## Execution
+
+Run this command via Bash, substituting the review request:
+
+```bash
+codex exec -c 'reasoning_effort="high"' "Review: $ARGUMENTS. Read all relevant files. Be specific — file paths, line numbers, exact issues. What bugs, edge cases, or design problems exist? Do NOT spawn sub-agents. Answer directly in bullet points." 2>&1
+```
+
+**Can run in background** (`run_in_background: true`) if you have other work to do in parallel.
+
+## Output
+
+Present the codex findings to the ape as:
+- Issues found (if any)
+- Agreements with your approach (if reviewing your own work)
+- Disagreements to address
+
+If codex found real issues in your code, fix them before presenting to the ape. Apes don't do tasks.