framework-evolution — 2026-04-19
Summary
2026-04-19 late-session: studied garrytan/gstack framework (YC-President-built, 23 specialists + 8 power tools + Builder Ethos + 10 host adapters, 522 files). Applied 8-anti-bloat-filter methodology — rejected 22 of 23 primitives (duplicate with Helix + Superpowers + our agents, OR Claude-only-stack-incompatible, OR scale-mismatch for solo CTO). Shipped 1 adoption: continuous-checkpoint.py PostToolUse hook for intra-session WIP persistence. 150-line Python hook, opt-in via marker/env, fires every 15 git-repo edits with structured [gstack-context] commit body. Unit tested: opt-out silent, opt-in fires at 15, non-git silent, file-count accurate. Cascade: hooks-registry.md (3 mentions), session-mcp-cleanup.py (2 new marker patterns), settings.json (registered under Write|Edit), session-learnings.md ([OPERATIONAL] entry with full rejection table). Test residue (15 leaked files from /root/cc-test-repo permission-denied test) caught + cleaned at session-close Phase 2.