feat: evolving memory + self-critique + coaching, bounded by token economy (3.3.0)

[Vahlame]

Makes the memory loop learn and improve service over time while staying token-cheap. The rules block (installed by --rules, plus install.md Step 4 and AGENTS.md) gains four tightly-written behaviors:

• Self-check before answering — assumptions stated? edge cases / failure modes? what would make this wrong? Scaled to the task, internal, no padding.
• Coach, don't impose — spot a high-impact anti-pattern in the user's code (secrets, unparameterized SQL, missing boundary types, force-push w/o lease, untested security rules)? Ask, and log a one-line hypothesis to PRACTICES/observations.md; promote to confirmed-{good,bad}.md only when the user confirms.
• Evolving memory — track new tech in STACKS/ (one note each), record firm preferences in MEMORY.md, hypotheses → facts, drop stale observations.
• Token economy (hard constraint) — passage-first reads, terse bullets, dedup. The intelligence comes from good notes + targeted recall, not re-reading or long monologues. So smarter ≠ pricier.

The scaffold now creates PRACTICES/ + STACKS/; examples/ gains matching samples. Rules kept dense (they load every session — keeping per-session cost low is part of the point). Package → 3.3.0.

Verified: npm test 24 + 41 · prettier 0 · markdownlint 0 · sync-agents OK · linkcheck 80.

🤖 Generated with Claude Code