docs(rules): add 'research before decisions' + 'deep-check before checkpoints' to standing rules

Per Lucas's direction 2026-05-08:
- Always do deep research / online research BEFORE making non-trivial
decisions (architecture, model picks, library choices, comp strategy).
Default-guessing is not allowed.
- Use deep-check skill for line-by-line whole-repo audits before any
submission, merge, or major checkpoint.

CLAUDE.md

@@ -9,6 +9,8 @@ Loaded when the working directory is inside `/Users/lucaslt/Documents/side-gig/a
 - **Never make assumptions — always look up answers online.** Before coding, configuring, or recommending anything, verify against authoritative sources (use `context7` for libraries / SDKs / APIs, `WebSearch` / `WebFetch` for everything else). Training data is stale; default-guesses waste time. This applies even to things that "seem obvious".
 - **Use Superpowers skills for every suitable use case — especially planning.** Any planning, debugging, executing-from-plan, brainstorming, parallel-agent dispatch, TDD, or pre-completion verification goes through the matching `superpowers:*` skill (`superpowers:writing-plans`, `:executing-plans`, `:brainstorming`, `:systematic-debugging`, `:subagent-driven-development`, `:verification-before-completion`, `:test-driven-development`, `:dispatching-parallel-agents`). Free-form prose plans are not allowed.
 - **Use the `deep-research` skill for deep academic research.** Multi-source comparison, literature review, state-of-the-art surveys, citation-tracked evidence — invoke `deep-research`, not ad-hoc web search.
+- **Always do deep research / online research BEFORE making non-trivial decisions.** Any architectural choice, model pick, library selection, or competition-strategy call goes through `deep-research` (academic) or `WebSearch` / `context7` (practical) first. Document findings inline so the decision is auditable. Default-guesses based on training data or "what feels right" are not allowed; the cost of looking things up is small, the cost of building on a wrong assumption is large.
+- **Use the `deep-check` skill for whole-repo audits before any submission, merge, or major checkpoint.** Run line-by-line bug + logic + security scan via `deep-check` after every meaningful change. Surface findings explicitly; fix blockers before declaring work done.
 
 ---