serjflint/saitenka

architecture-inquiry

Prepare a grounded architecture decision across interacting operational slices — a Bohm-inspired whole-to-parts-to-whole inquiry for god objects, ownership boundaries, architecture smell, event/reducer/controller choices, and decision dossiers. Starts from…

原文语言：英语

architecture-review

Judge whether the existing architecture is fit for this product through an isolated, code-grounded reviewer and explicit quality axes. Use after a migration, before a feature that assumes the current shape, when the meters are green but something feels wrong,…

原文语言：英语

plan-migration

Price an already-chosen migration before it starts: enumerate the class, census and price the unit, choose and reject leverage devices explicitly, declare retirement beside debt, then batch the family and codemod the mechanical part. Use for a…

原文语言：英语

contribute

Drive a change big enough to warrant its own PR — from diagnosis to a review-ready pull request — the house way: the ordered loop this repo and its upstream forks (repowise) use: restore context, diagnose to root cause offline, PoC to prove tractability,…

原文语言：英语

write-test

Author a new pytest test for the Saitenka overlay the house way — pick the test tier, decide real-vs-fake at each boundary, construct-vs-fixture, and the assertion ladder, then write it against the real fakes. Use when adding or rewriting a test under tests/:…

原文语言：英语

research

Run grounded prior-art, SOTA, or competitor research where an external system proposes and we verify against fetched sources and our repo/memory. Use for “research prior art”, “survey the field”, “compare approaches”, “is X state of the art”, or a rival…

原文语言：英语

sharpen-loop

Run or resume one Saitenka Sharpen audit: triage a module, measure its existing tests, propose a test-only improvement, apply the deterministic anti-lobotomization gate, obtain two isolated adversarial reviews, and record a dry-run or PR-ready result. Trigger…

原文语言：英语

skill-creator

Author a new saitenka .agents/skills skill well, once a skill is the right home — the structure (SKILL.md procedure + references/ depth + grep-free scripts/smoke.sh + optional agents/openai.yaml), a discovery-optimized trigger description, progressive…

原文语言：英语

pyrefly-lsp

Do not invoke. Infrastructure-only — the sibling `.lsp.json` registers the pyrefly Python language server (`pyrefly lsp`) with Claude Code's built-in LSP tool, giving go-to-definition, find-references, call hierarchy, and document symbols for .py/.pyi files.…

原文语言：英语

agents-md

Decide where an agent-facing convention or constraint belongs, and keep AGENTS.md minimal — routing across saitenka's config surfaces (AGENTS.md, .agents/rules/, .agents/skills/, .agents/hooks/, poe required-vs-advisory gates, .agents/mcp/servers.json). Use…

原文语言：英语

dev-gate

Run and interpret the Saitenka quality gates — the fast pre-push `poe all` and the slower pre-tag `poe pre-release`. Use before pushing or when asked to "run the gate", "run poe all", "check it passes / is it green", "lint/types/arch/tests before commit",…

原文语言：英语

grow-loop

Run or resume one Saitenka Grow audit: find an orphan scenario, choose the test kind and oracle through write-test, add one focused test, run the deterministic growth gate, obtain two isolated adversarial reviews, and record a dry-run or PR-ready result.…

原文语言：英语

test-adequacy

Run and interpret saitenka's opt-in test-adequacy tools for the pure core — mutation (`poe mutate`, cosmic-ray), coverage-guided fuzzing (`poe fuzz`, atheris), symbolic execution (`poe crosshair`, z3) — and harden what they find. Use when asked to "run…

原文语言：英语

agent-tooling

Pick the right code-intelligence tool for a task in this repo — repowise (MCP), the pyrefly LSP tool, or Basic Memory (MCP). Use when deciding how to answer "who calls / where is / find the symbol", "what's risky to change", "what did we decide and why",…

原文语言：英语

agent-setup

Enable this repo's agent tooling (repowise + Basic Memory MCP, pyrefly LSP, repowise hooks) in the coding agent you use — Claude Code or Codex. Use when asked to "set up the tooling", "wire up repowise / Basic Memory", "enable the MCP servers", "register the…

原文语言：英语

dream

Refactor this project's local, mutable Claude memory (the per-fact files plus the MEMORY.md index under the project's ~/.claude memory dir) so it stays a lean, accurate, always-loaded index — and uplift the depth into Basic Memory (the deep durable store, MCP…

原文语言：英语