TT Claude Memory

feedback_5_step_phase_0_hardening_cycle

Properties1
descriptionThe full Phase-0 hardening cycle is Session 1 → Session 2 → CLAUDE.md cross-pollination → /review-plan + doc-reconciliation → second /review-plan verification + 5-min follow-up; 2 of 5 v2 build repos are at full 5-step parity

3 min read

The full Phase-0 hardening cycle for a TT brand build repo, as proven across 5 v2 build repos on 2026-04-27 → 2026-04-28:

  1. Session 1 (legal + tooling foundation): LICENSE + NOTICE + THIRD-PARTY-DEPENDENCIES.md + README.md + package.json/pyproject.toml + tsconfig.json + .gitignore + .editorconfig + .prettierrc + .github/workflows/{license-audit, ci}.yml. Scope: Apache 2.0 / AGPL+commercial / commercial-proprietary licensing per tier (Tier-3 / Tier-2 / Tier-1) + appropriate misalignment guards in license-audit workflow.

  2. Session 2 (docs + scaffolding + BUILD-PLAN.md v1.1): folder skeleton + stub READMEs + docs/ci-pipeline.md operator runbook + examples/first-fixture/ orientation example + BUILD-PLAN.md v1.0 → v1.1 hardening (§1.4 Day-1 onboarding 5-sub-section + §3.0 Phase-0 substance + §3.4 per-stage detail + §3.5 receives-from-each-build-plan + §3.6 19-CI-gate enumeration).

  3. CLAUDE.md cross-pollination: project-level CLAUDE.md authored per the 4-repo template (§1 locked conventions table + §2 reading order + §3 drift-prevention discipline + §4-§N tier-specific + §contact). Active-work-log scaffolded per global CLAUDE.md §7 parallel-sessions discipline.

  4. /review-plan + doc-reconciliation: invoke /review-plan skill; ~13 findings per repo (consistent 3 HIGH + 5 MEDIUM + 5 LOW shape across both repos that have completed this step); reconcile all in single commit per CLAUDE.md §3 back-propagate-in-same-commit-cycle discipline. Bump BUILD-PLAN.md v1.1 → v1.2.

  5. Second /review-plan verification + 5-min follow-up: invoke /review-plan again; ~5x cheaper (verification not discovery); typically catches 2 new minor issues introduced by the reconciliation itself (banner-sweep gaps, etc.); single follow-up commit clears them.

Why: Reproducible across 5 distinct repository archetypes (Tier-3 Apache 2.0 standard / Tier-3 Apache 2.0 website / Tier-2 AGPL+commercial dual / Tier-1 commercial-proprietary single / cross-cutting Apache 2.0 dual-stack TS+Python). Total wall-clock per repo through full 5 steps: ~5-6 hours. Total commits per repo: ~7-10 across the build repo + docs-strategy combined. The 5-step cycle is now the new gold standard for Phase-0 hardening of TT brand build repos.

Status as of 2026-04-28T19:45:

RepoTierSteps completeFinal state
code-inherit-v2Tier-3 standard (Apache 2.0)1+2+3 (no /review-plan yet)BUILD-PLAN v1.1, CLAUDE.md, no reconciliation
code-inheritkitTier-2 SDK (AGPL+commercial)1+2+3 (no /review-plan yet)BUILD-PLAN v1.1, CLAUDE.md, no reconciliation
code-iasTier-1 IAS (commercial-proprietary)1+2+3 (no /review-plan yet)BUILD-PLAN v1.1, CLAUDE.md, no reconciliation
code-inheritv2-wwwTier-3 website (Apache 2.0)1+2+3+4BUILD-PLAN v1.2, CLAUDE.md, /review-plan done; no second-pass verification
code-inheritv2-test-suiteCross-cutting test-pyramid (Apache 2.0; dual-stack TS+Python)1+2+3+4+5BUILD-PLAN v1.2, CLAUDE.md, both /review-plan passes complete

How to apply:

For any future TT brand build repo (e.g., code-inheritwills + code-myfamilyinherits + code-ll-v2 per brand-architecture v1.4):

  1. Run all 5 steps in sequence; do not skip any
  2. Each step has its own autopilot prompt template — adapt from the v2-test-suite Session 1 / Session 2 templates
  3. Total budget per repo: ~5-6 hours wall-clock (Session 1 ~2-3h + Session 2 ~3-4h + CLAUDE.md ~15min + /review-plan reconciliation ~30min + 5-min follow-up ~15min)
  4. Forward candidates for 4-step → 5-step parity on existing v2 repos: code-inherit-v2 + code-inheritkit + code-ias each have only 1+2+3 complete; could bring all 5 to full 5-step parity in ~2-2.5 hours of /review-plan reconciliation passes

Graph View

Backlinks