cheese-flow

Staff Engineer code review. Runs nine orthogonal LLM dimensions (correctness, security, complexity, encapsulation, spec, precedent, deslop, assertions, nih) over the diff and emits a stake-weighted report plus hash-anchored sidecar JSON consumed by /cleanup and /fromage cook.

Multi-source research orchestrator. Routes a question across Tavily, Context7 library docs, GitHub, and in-repo tools, writes findings to .cheese/research/<slug>.md, and returns a compact synthesis to the main context.

Unified entry point that inspects user input, announces detected intent, and routes to the right downstream skill (/mold, /cook, /age, /briesearch, or a debug/review flow) before dispatching.

Sequential spec-to-implementation flow that runs cut, cook, taste-test, press, and (optionally) assertion-review using Cheez skills as the backbone.

Free-form rubber-ducking and architecture exploration. Hard invariant is no writes to production files; the goal is a shared mental model, not code, spec, or PR output.

Finish what /age started — load both sidecars, render a unified stake table, gate on user approval, apply via /cleanup + cook sub-agents, then re-age the touched paths up to a 3-turn cap.

Iterative thinking amplifier for fuzzy ideas. Routes input to a starting mode (Explore, Ground, Shape, Sketch, Grill, Diagnose), runs Validate Cycles, locks down interfaces in pseudocode, and curdles a spec (and optional issues) only after a two-key handshake.

Whole-repo audit for code that reinvents what open-source libraries already do. Detects hand-rolled retry, validation, UUID, debounce, date, argparse, and other common patterns; cross-checks against installed dependencies; returns scored migration recommendations with effort estimates.

Staff Engineer code review orchestrator. Runs nine orthogonal LLM dimensions over a diff and emits a stake-weighted report plus hash-anchored sidecar JSON.

A portable starter skill in Agent Skills format for repositories adopting cheese-flow.

Smart code reading using tilth MCP. Replaces cat/head/tail with AST-aware file reading. Automatically outlines large files, memorizes hash anchors for efficient editing, and tracks what you've already read to save tokens. Use when: reading files, exploring code structure, understanding modules, checking dependencies, or preparing for edits. Do NOT use for searching symbols or text — use cheez-search. Do NOT use for editing — use cheez-write. Examples: "read src/auth.ts", "show lines 44-89 of handlers.go", "what's in this directory?", "show the dependencies of this file".

AST-aware code search using tilth MCP. Finds definitions first, then usages. Use tree-sitter structural matching instead of blind text grep. Understand dependencies and call graphs. Use when: finding where symbols are defined, tracing call chains, understanding code structure, finding all callers of a function. Do NOT use for reading files — use cheez-read. Do NOT use for editing — use cheez-write. Examples: "find where handleAuth is defined", "what calls validateToken?", "trace the ServeHTTP, HandlersChain, Next functions", "find all implementations of the UserService interface".

Hash-anchored file editing using tilth MCP. Replace specific functions or blocks using hash anchors from tilth_read, not entire file rewrites. Use when: editing code, replacing functions, deleting blocks, or making precise surgical changes. Always read first to get hash anchors. Do NOT use for reading files — use cheez-read. Do NOT use for searching — use cheez-search. Examples: "replace the validateToken function with this new implementation", "delete lines 44-89", "update the authenticate method".

Mechanical apply of /age's hash-anchored sidecar fixes via tilth_edit. The cleanup-wolf sub-agent is the only LLM in the path, fired per anchor mismatch.

Finish what /age started. Loads both sidecars (fixes + suggestions), renders a unified stake table, gates on user approval, routes each approved item to the right handler (/cleanup or a cook sub-agent), then re-ages the touched paths up to a hard 3-turn cap.

Complete GitHub tasks using the gh CLI. Use for any GitHub operation — PRs, issues, CI checks, repo management, releases, code search. Use git commands (log, diff, status) for local context. Use when the user says "create PR", "merge PR", "check CI", "list issues", "review PR", "PR status", "close issue". Do NOT use for local git operations like commit, stage, or push — use commit skill for those. Do NOT use for code quality review — use age or code-review. Examples: "review PR 14", "create a PR for my branch", "what's the CI status on this PR?", "list open issues labeled bug", "merge PR 23 with squash", "show me issue 42 with comments".

Resolve merge conflicts, rebase conflicts, and cherry-pick failures using mergiraf (AST-aware structural merge), git rerere, and kdiff3. Activate when: merge failed, rebase conflict, cherry-pick failed, CONFLICT in file, "resolve conflicts", "fix merge", "merge conflict", "conflict resolution", or git output shows CONFLICT markers. Also use for mergiraf diagnostics, rerere management, gitattributes regeneration, or batch conflict resolution across multiple files. Covers the full resolution chain: mergiraf (structural auto-resolve) → rerere (replay remembered fixes) → kdiff3 (manual). Do NOT use for general git operations without conflicts — those go to the commit or gh skills. Examples: "resolve the merge conflicts", "fix the rebase conflicts", "what's conflicting after the merge?", "resolve conflicts in src/auth.ts".

Use the Milknado MCP tools to pick and execute the next ready node from the Mikado graph, apply file changes, and run quality gates.

Use the Milknado MCP tools to decompose a goal into a Mikado-style change graph and produce a token-budgeted batch plan.

Iterative thinking amplifier for fuzzy ideas. Routes input to the right starting mode (Explore, Ground, Shape, Sketch, Grill, Diagnose), runs Validate Cycles to anchor claims, locks down interfaces in pseudocode, and only curdles a spec (and optional issues) after a two-key handshake plus a coherence self-check.

Scan a codebase for custom code that duplicates what open-source libraries already do, then recommend which libraries to adopt. Detects hand-rolled utility functions, custom retry logic, manual validation, DIY date handling, home-grown argument parsers, and other reinvented wheels. Cross-checks against installed dependencies and open specs. Returns scored migration recommendations with effort estimates. Use when the user mentions reinventing the wheel, asks if there is a library for something they built, wants a build-vs-buy audit, asks "should we just use lodash for this", or wants to find dependency opportunities.

Multi-source research orchestrator that routes questions across Context7 library docs, Tavily web research, codebase analysis, and GitHub examples, then returns a compact synthesis with confidence scoring.