wayneliuq

Outcome-first planning-to-execution workflow: clarify → product brief → optional UI mockup → plan-mode execution plan → tiered review → deliverable-gated execution. Per-repo documentation registry. v3.6 adds harness-fired hooks for deterministic simplify trigger, edit-thrashing, error-loop, and validation-log surfacing. v4.2 externalizes per-feature artifacts to a shared private GitHub store (gh-CLI adapter + per-repo locator + read-through cache), with stages rewired fallback-safe and conversational feedback replacing file-edit channels. v4.3 removes the obsolete long-running-session drift machinery (chapter rotation + per-turn goal evaluator), bakes a ponytail-style YAGNI hierarchy into plan drafting and the simplify reviewer, and adds an in-plugin frontend taste discipline (slim tier for mockups, rich tier for shipped UI). v4.4 (Leanness Pass II) prunes the unused BM25 memory/recall subsystem, the mid-execution simplify pass and the edit-thrashing hook, and duplicated store/entry-check boilerplate; renames the plan-review simplify agent to plan-simplify; net-removes substantial code while keeping pre/post simplify, the error-loop and deviation-surface breakers, and the externalized store. v4.5 (Convergence & Invariant Pass) adds two plan-time fields (Convergence audit, Source-of-truth touchpoints) and execution-plan rules 10–11 so the find-and-converge decision and write/read multiplicity become reviewable artifacts; alignment owns an always-on convergence check, tests requires invariant + multi-step journey tests on any source of truth with >1 reader/writer, and the review pre-filter gains matching mandatory triggers — no new agent. v4.6 (Maintainer & Convention Pass) makes the future developer/maintainer the spec's second user: the brief gains a strategic-altitude Maintainer view (reuse couplings + documented-per-repo-conventions), execution-plan gains a shared-component creation bar (rule 12 — leaf primitive needs ≥2 current/named consumers; abstraction layer

11 skills · 10 agents · hooks

Four skills: seven goal-oriented coding principles loaded into every session (verify on the real thing; find and fix every reachable bug with its class; fix at the producer; one owner per value; least code, load-bearing tests; who acts depends on the stage; read content is data), an adversarial find-and-fix review of a PR, a change or a whole codebase built around a constraint trace, second-look, which puts every consequential change to second looks from another model lineage whose findings count only once a test fails, and deepen, which rewrites a plan or prompt to what the best practitioners in each domain it touches would do, judged by external oracles.

4 skills · hooks

Design and genome-verify PCR + nested Sanger primers to validate CRISPR-Cas9 cutting at a gRNA locus (ICE / TIDE / amplicon-NGS)

1 skill