π§ easy-cheese π§
Don't know what to do? Just /cheese it.
"The cheese must flow."
A portable, harness-agnostic Agent Skills toolkit β self-contained SKILL.md files any Agent Skills-compatible harness can load. No agents, no compiled bundles, no repo-wide MCP requirement. The vocabulary (mold, culture, cook, press, age, cure) reads as a workflow you can dip into anywhere.
Contents
- Why cheese?
- Skill layout
- Skills
- Scope
- Optional tools
- Install
- Validate
- Installing MCP servers
- Installing CLI tools
- Credits
Why cheese? Two reasons
- From gaming slang: a "cheese" win is cheap, easy, and disproportionately effective β exactly the design center (correctness, token efficiency, quality).
- What's life without whimsy? π§
Skill layout
This repo follows the Agent Skills spec:
skills/
βββ <skill-name>/
βββ SKILL.md # required: name + description + body
βββ references/ # optional: detail pulled in on demand
βββ scripts/ # optional: executable helpers
βββ assets/ # optional: templates / static resources
Each SKILL.md is self-contained markdown with YAML frontmatter. There are no nested sub-skills; deeper material lives in references/<topic>.md so the harness can load it progressively.
Content shared across skills lives in the always-installed cheese skill's references/ directory (e.g. skills/cheese/references/handoff-gate.md), so both installers ship it as part of cheese. Skills reference it by sibling-relative path (../cheese/references/<file>.md from a SKILL.md, ../../cheese/references/<file>.md from a references/*.md) β a form that resolves identically in the repo tree and any installed skills directory.
Skills
Workflow skills
| Skill path | Command | Purpose |
|---|---|---|
skills/cheese/SKILL.md |
/cheese |
Unified entry point. Classifies any input (idea, spec path, PR, stack trace, file path), announces the routing decision as a short three-line block (Intent / Reason / Target), and dispatches the chosen target immediately with --auto propagated downstream. Add --safe to gate dispatch behind a confirmation prompt and surface non-auto alternatives. |
skills/briesearch/SKILL.md |
/briesearch |
Research technical questions across docs, web, codebase, and GitHub examples with confidence-capped synthesis. |
skills/mold/SKILL.md |
/mold |
Shape fuzzy ideas into grounded specs through dialogue, validate cycles, and a two-key handshake. |
skills/culture/SKILL.md |
/culture |
The agent's internal-thinking skill β invoked silently by /cheese and other workflow skills to model a problem before dispatching. Surfaces to the user only when they explicitly opted out of writes ("no writes", "rubber-duck this"). Hard invariant: writes only the opt-in .cheese/notes/<slug>.md handoff at session end, and only when the user asks for notes. |
skills/pasteurize/SKILL.md |
/pasteurize |
Diagnose hard bugs, flaky failures, and performance regressions with a feedback-loop-first investigation, then hand off into /cook β /press β /age β /cure. |
skills/cook/SKILL.md |
/cook |
Implement clear specs via cut β cook β taste-test with scoped edits and tests. |
skills/press/SKILL.md |
/press |
Harden cooked changes with coverage, assertion, and boundary checks. |
skills/age/SKILL.md |
/age |
Review diffs across ten staff-engineer dimensions and produce a severity-grouped findings report. |
skills/affinage/SKILL.md |
/affinage |
Triage external PR claims β review comments and CI failures β through the /age lens, hand the chosen fixes to /cure, then post replies back on GitHub. |
skills/cure/SKILL.md |
/cure |
Fix user-selected findings, validate, and prepare the branch for shipping. |
skills/plate/SKILL.md |
/plate |
Final writing, validation, commit, and ordinary-or-stacked PR publication gate. Explicit topology choices win; obvious cohesive work opens as one PR, while stack recommendations and ambiguous shapes ask before mutation. |
skills/hard-cheese/SKILL.md |
/hard-cheese (or --hard flag) |
Metacognitive vibecheck gate before review β asks the author to explain the diff's causal logic, grades the explanation against the SOLO Taxonomy. Standalone or via --hard propagation through the pipeline. |
skills/ultracook/SKILL.md |
/ultracook |
Autonomous fresh-context pipeline for high-blast-radius specs. Typed phase agents run each curd as sequential same-worktree spawns: coder (cook) β coder (press) β reviewer (age) β coder (cure) β reviewer (final age), followed by parent-owned commit-only /plate. Post-merge uses press β age β cure β age; an indivisible spec keeps the seven-phase linear chain. |
skills/melt/SKILL.md |
/melt |
Resolve merge / rebase / cherry-pick conflicts via the structural cascade (mergiraf β rerere β kdiff3) with batch, pick-side, and lockfile helpers. |
skills/wheypoint/SKILL.md |
/wheypoint |
Mark a checkpoint: compact a mid-task conversation into a durable handoff document at .cheese/notes/<slug>.md (resumable slug + state-mapped suggested-skills + redacted secrets) so a fresh agent can resume via /cheese --continue <slug>. |
Tool skills
The workflow skills can delegate code search, reading, and editing to these backend-aware skills when tilth or an equivalent native source-code backend is available:
| Skill path | Command | Purpose |
|---|---|---|
skills/cheez-search/SKILL.md |
/cheez-search |
AST-aware code/text/regex/caller search via tilth MCP, harness-native AST search, or LSP. Replaces grep / rg / find. |
skills/cheez-read/SKILL.md |
/cheez-read |
Smart file/directory reading through a freshness-aware backend; tilth MCP is preferred because it emits stale-write-safe edit tags. Replaces cat / head / tail / ls. |
skills/cheez-write/SKILL.md |
/cheez-write |
Anchored, surgical edits through tilth MCP or an equivalent harness-native stale-snapshot edit tool. Never rewrites whole files blindly. |
Dispatching skills resolve agent type, power, permissions, isolation, and fallback through the shared agent-resolution.md contract.
The cheez-* skills tell the model which source-code backend to use when the harness has one. Prefer tilth MCP by name; use LSP for type-grounded definitions/references/renames/code actions, ast_grep/sg for structural patterns and codemods, and anchored/stale-checking edits for ordinary block changes. Plain text shell tools are fallback evidence only, not equivalent source-code backends.
cheez-* router protocol
The three cheez-* skills are designed to chain. The standard sequence:
/cheez-searchβ locate the symbol, caller, content match, or file with tilth, AST search, or LSP./cheez-readβ read the target file or section with enough line/hash/snapshot context for a safe edit./cheez-writeβ apply the change through tilth edit, LSP workspace edit, AST rewrite, or another stale-checking edit backend.
Workflow skills (/cook, /age, /cure) call into this chain when they need code intelligence. A skill should never search-then-edit without reading in between β the read is what produces the anchors that make the edit safe.
Source-code tool routing
For source code, route by capability instead of command name:
- LSP wins for type-grounded definitions, references, renames, and code actions.
- AST search /
sgwins for syntax-shaped patterns and codemods. - tilth wins for broad source search/read/edit context, anchors, outlines, token estimates, and dependency context.
- Anchored or stale-checking edits win for ordinary block/range changes.
Plain shell search/view/edit is fallback evidence only; use it for non-code paths, logs, generated output, or when no semantic backend exists, and name the downgrade.
Installing tilth MCP
See Installing MCP servers below β expand the tilth section for the preferred cheez-* backend. If tilth is unavailable, use equivalent native AST/LSP/anchored-edit tooling when the harness provides it; otherwise name the weaker fallback instead of pretending blind shell tools have the same safety.
Suggested flow
/cheese βββΊ classify intent βββΊ dispatch immediately (autonomous by default)
ββ need info / external evidence βββΊ /briesearch
ββ no-writes discussion only βββΊ /culture (user explicitly opted out of writes)
ββ fuzzy / multi-module idea βββΊ /mold βββΊ /cook --auto βββΊ /press βββΊ /age βββΊ /cure
ββ high-blast-radius spec βββΊ /mold βββΊ /ultracook (fresh-context: cook β press β age β cure β age β cure β age)
ββ clear, scoped ask βββΊ /cook --auto βββΊ /press βββΊ /age βββΊ /cure
ββ debugging task βββΊ /pasteurize --auto βββΊ /cook --auto βββΊ /press βββΊ /age βββΊ /cure
ββ PR comments / CI failures βββΊ /affinage βββΊ /cure
ββ running low on context βββΊ /wheypoint βββΊ /cheese --continue <slug> (fresh session)
ββ resume in fresh context βββΊ /cheese --continue <slug>
ββ review only βββΊ /age βββΊ /cure
/cheese is the front door. It inspects whatever you drop in (idea, spec path,
PR ref, stack trace, file path), announces its routing decision as a short
three-line block (Intent / Reason / Target), and dispatches the chosen skill in
the same turn β --auto propagates
downstream so the chain runs all the way through. Use --safe when you want
the chance to redirect before anything runs: it puts the confirmation prompt
back in front of dispatch and surfaces non-auto variants as alternatives. Skip
/cheese entirely when you already know the destination β a hard bug can go
straight to /pasteurize, a known-scope fix can go to /cook, and an
explicit no-writes design discussion goes to /culture. /melt cuts in
whenever a merge step blocks /cook or /cure. Append --hard to any
pipeline step to insert /hard-cheese as a metacognitive vibecheck gate
before review.
Scope
Easy-cheese is intentionally a small surface. What that means in practice:
- Skills only. No agents, commands, eta templates, or compiled harness bundles. Each capability is a single
SKILL.md. - No repo-wide MCP requirement. Workflow skills suggest tools (tilth, Context7, Tavily) but have host-native fallbacks. The cheez-* tool skills require an AST-aware search/read path and stale-checking edit path; prefer tilth when present, but equivalent native AST/LSP/anchored-edit backends satisfy the contract.
- One orchestrator skill, narrowly scoped.
/ultracookis the only orchestrator. It resolves planner/coder/reviewer phase agents by capability and minimum power; harvest and/plateremain parent-owned. Parallel curds use sequential same-worktree phase spawns and a terminal reviewer pass before publication. - No automatic re-age loop in
/cure. The skill describes the protocol; the human runs the next/agewhen ready.
Optional tools
Workflow skills name preferred tools when they help, with fallbacks for portability. Tool skills can be stricter when their purpose is to enforce a specific tool protocol.
| Tool | Helps with | Fallback |
|---|---|---|
| tilth (MCP) | AST-aware read/search/edit and dependency context | Preferred for cheez-*; native AST/LSP/anchored-edit tools may satisfy the same contract |
sg (ast-grep) |
Structural pattern matching and codemods (sg --rewrite) with metavariables |
LSP or tilth for symbol/caller questions; anchored edits for non-structural block changes |
| Context7 (MCP) | Library and API documentation | repo docs, package docs, vendor pages, web search |
| Tavily (MCP) | Current web/vendor research | host web search or user-supplied sources |
| LSP / Serena (MCP) | Type-aware xrefs (find_referencing_symbols, find_implementations), symbol-bounded edits (rename_symbol, replace_symbol_body, safe_delete_symbol), and LSP diagnostics β concrete tools for the abstract "if your harness has an LSP" sections in cheez-* skills |
sg, tilth_search, targeted reads via tilth |
| hallouminate (MCP) | Per-repo wiki for cross-session design rationale, ADR grounding, and /mold evidence |
Skip wiki grounding; proceed with diff + code evidence only; cap at speculating when design rationale is central |
| milknado (MCP) | Mikado task-graph backend for /ultracook parallel-mode curd prerequisite tracking |
In-report curd decomposition in manifest YAML; no external task-graph backend needed |
ripgrep |
Fast text search | grep, find, editor search |
gh |
GitHub issues, PRs, checks, examples | local git commands or user-provided links/logs |
delta |
Readable diffs | plain git diff |
mergiraf |
Structured merge conflict resolution | manual conflict resolution plus tests |
jq |
JSON inspection for reports or tool output | manual inspection |
fd |
Fast file discovery | find |
just |
Project task discovery | package scripts or documented commands |
When a preferred tool is unavailable, workflow skills say so once, fall back, and lower confidence only if evidence quality suffers. The cheez-* skills should teach the strongest available backend shape, not break on a missing MCP.
Install
skills.sh (recommended)
Install every skill with the skills.sh installer. The
skills form one pipeline (/cheese β /mold β /cook β /press β /age β /cure),
so install them together rather than cherry-picking:
npx skills@latest add paulnsorensen/easy-cheese --all --global
--all is shorthand for --skill "*" --agent "*" -y: every published skill,
into every detected agent, no prompts. --global installs them user-wide, so
they are available across every repo.
To scope the install to just the current repo instead (recorded under
.agents/skills/), drop --global and run it from that repo:
npx skills@latest add paulnsorensen/easy-cheese --all
After install, start with /cheese if you're not sure which wheel to cut into
first, or jump straight to a specific skill like /cook, /age, or
/pasteurize. There is no required follow-up setup skill.
gh skill (GitHub CLI alternative)
Requires GitHub CLI v2.90.0 or later with the gh skill command.
Install all skills interactively:
gh skill install paulnsorensen/easy-cheese
Install every current skill in one shot:
for s in age affinage briesearch cheese cheez-read cheez-search cheez-write cook culture cure hard-cheese melt mold pasteurize press ultracook wheypoint; do
gh skill install paulnsorensen/easy-cheese "$s"
done
Install one specific skill by name:
gh skill install paulnsorensen/easy-cheese cook
Pin to a specific release tag or commit SHA for reproducibility:
gh skill install paulnsorensen/easy-cheese [email protected]
gh skill install paulnsorensen/easy-cheese cook@abc123def
Control which agent and scope to install into:
# User-wide (recommended for personal toolkits)
gh skill install paulnsorensen/easy-cheese --agent claude-code --scope user
# Committed into the current project repo (default scope)
gh skill install paulnsorensen/easy-cheese --agent claude-code --scope project
Supported --agent values include github-copilot, claude-code, cursor, codex, gemini-cli, and others. Omit --agent to use the harness auto-detected from your environment.
Preview a skill's content before committing to an install:
gh skill preview paulnsorensen/easy-cheese cook
Keep installed skills up to date:
gh skill update --all
Claude Code (manual)
Copy the skills you want into your skills directory:
# Per-user
mkdir -p ~/.claude/skills
cp -r skills/age ~/.claude/skills/
# Per-project
mkdir -p .claude/skills
cp -r skills/cook .claude/skills/
Other harnesses
Copy skills/<name>/ into wherever the harness loads Agent Skills from. The format follows the agentskills.io spec and works in any compliant client.
Validate
The reference validator from agentskills/agentskills checks frontmatter and naming:
pip install skills-ref # ships the `agentskills` CLI
agentskills validate ./skills/age
Each SKILL.md must have YAML frontmatter with at least name and description, and name must match the parent directory name.
Installing MCP servers
The cheez-* tool skills and several workflow skills benefit from MCP servers. Install the ones you need.
tilth provides AST-aware code search, smart file reading, and tag-anchored edits. It is the preferred backend for /cheez-search, /cheez-read, and /cheez-write; equivalent native AST/LSP/anchored-edit backends satisfy the same contract when Tilth is unavailable.
# Install tilth CLI via npm (Node.js v18+) β no Homebrew formula upstream
npm install -g @paulnsorensen/tilth-nightly@latest
# or run via npx β no global install needed:
# npx -y @paulnsorensen/tilth-nightly@latest install claude-code --edit
# Register as an MCP server β include --edit only if you plan to use cheez-write
tilth install claude-code --edit # Claude Code
tilth install cursor --edit # Cursor
tilth install vscode --edit # VS Code
tilth install codex --edit # Codex CLI
tilth install gemini --edit # Gemini CLI
tilth install zed --edit # Zed
After registering, restart your harness and confirm these tools appear:
mcp__tilth__tilth_searchmcp__tilth__tilth_readmcp__tilth__tilth_listmcp__tilth__tilth_depsmcp__tilth__tilth_write(only with--edit)
Context7 fetches up-to-date, version-specific library docs into your session. Used by /briesearch and /cook when available.
Claude Code:
claude mcp add context7 -- npx -y @upstash/context7-mcp@latest
Other harnesses β add to your MCP config file:
{
"mcpServers": {
"context7": {
"command": "npx",
"args": ["-y", "@upstash/context7-mcp@latest"]
}
}
}
For higher rate limits, get a free API key at context7.com and append --api-key YOUR_API_KEY to the args array. A keyless hosted option is also available:
{
"mcpServers": {
"context7": {
"url": "https://mcp.context7.com/mcp"
}
}
}
Requires Node.js v18+.
Tavily provides real-time web search and content extraction. Used by /briesearch when available.
Get a free API key at tavily.com, then:
Claude Code:
claude mcp add tavily -- npx -y tavily-mcp
Set your key in the environment or pass it inline:
TAVILY_API_KEY=your-key npx -y tavily-mcp
Other harnesses β add to your MCP config file:
{
"mcpServers": {
"tavily": {
"command": "npx",
"args": ["-y", "tavily-mcp"],
"env": {
"TAVILY_API_KEY": "your-key"
}
}
}
}
Requires Node.js v18+.
Installing CLI tools
The optional tools listed under Optional tools are referenced by workflow skills. None are required, but having them available unlocks better fallbacks and richer output.
macOS bootstrap script (optional)
Use
scripts/install.sh
when you want the surrounding macOS toolchain and MCP servers set up for you.
The recommended way to install the skills themselves is still the skills.sh
flow above; this script is the fast lane for the wider ecosystem.
It does the following in one shot:
- Installs every CLI tool listed below β Homebrew for the eight brew-core formulas, plus
npm install -g @paulnsorensen/tilth-nightly@latestfor tilth, which has no Homebrew formula upstream. - Auto-detects installed Claude Code, Cursor, and Codex CLIs, then installs every easy-cheese skill into each detected harness at user scope as a convenience bootstrap.
- Registers the
tilth,context7, andhallouminateMCP servers with those harnesses where supported. hallouminate is delivered as a plugin: onclaude-codeandcodexthe installer adds thepaulnsorensen/hallouminateplugin marketplace and installs the plugin (which ships its own MCP server and bootstraps the binary); on other harnesses it warns and you register the MCP manually. Other servers (tavily,milknado) are also selectable with--mcp.
Currently macOS only β it relies on Homebrew. Requires gh to be authenticated (gh auth login) before running.
Pipe straight from GitHub:
curl -fsSL https://raw.githubusercontent.com/paulnsorensen/easy-cheese/main/scripts/install.sh | bash
Or grab the script first if you'd like to read it:
curl -fsSL -o /tmp/easy-cheese-install.sh https://raw.githubusercontent.com/paulnsorensen/easy-cheese/main/scripts/install.sh
bash /tmp/easy-cheese-install.sh --help
bash /tmp/easy-cheese-install.sh --dry-run
Common flags:
# Install only ripgrep + jq, skip MCP registration
curl -fsSL https://raw.githubusercontent.com/paulnsorensen/easy-cheese/main/scripts/install.sh \
| bash -s -- --tools ripgrep,jq --skip-mcp
# Register MCP servers only (assumes CLI tools and skills are already installed)
curl -fsSL https://raw.githubusercontent.com/paulnsorensen/easy-cheese/main/scripts/install.sh \
| bash -s -- --skip-tools --mcp tilth,context7,tavily
# Pick a specific harness for skill + MCP registration
curl -fsSL https://raw.githubusercontent.com/paulnsorensen/easy-cheese/main/scripts/install.sh \
| bash -s -- --harness cursor
# Or target a comma-separated harness list explicitly
curl -fsSL https://raw.githubusercontent.com/paulnsorensen/easy-cheese/main/scripts/install.sh \
| bash -s -- --harness claude-code,cursor,codex
The script is idempotent β it skips any tool already on PATH β and accepts --dry-run so you can preview what it would do before letting it run. If no supported harness CLI is detected, it falls back to the historical claude-code target; pass --harness to override detection.
Heads-up:
curl | bashruns whatever the URL serves at the moment of the request. If you want to audit before running, use the two-step form above.
If you'd rather install tools individually, the per-tool sections below cover macOS, Windows, and Linux.
GitHub CLI (gh)
brew install gh # macOS/Linux via Homebrew
winget install GitHub.cli # Windows
# or see https://cli.github.com for other methods
gh auth login
Minimum version for gh skill: v2.90.0.
gh --version
gh skill ships as a built-in subcommand in GitHub CLI v2.90.0+. If your installation predates that release, upgrade gh rather than installing an extension. Check cli.github.com/manual/gh_skill for the current status.
ast-grep (sg)
Used by /cook, /age, and /cure for structural codemods when tilth is unavailable.
brew install ast-grep # macOS/Linux
npm install -g @ast-grep/cli # Node.js
cargo install ast-grep # Rust/Cargo
scoop install ast-grep # Windows (Scoop)
ripgrep (rg)
Fast text search used as a fallback when tilth is unavailable.
brew install ripgrep # macOS/Linux
winget install BurntSushi.ripgrep.MSVC # Windows
cargo install ripgrep # Rust/Cargo
delta
Human-readable diffs used by /age and /cure.
brew install git-delta # macOS/Linux
cargo install git-delta # Rust/Cargo
winget install dandavison.delta # Windows
Add to ~/.gitconfig to enable globally:
[core]
pager = delta
[interactive]
diffFilter = delta --color-only
mergiraf
Structured merge-conflict resolution used by /melt.
cargo install mergiraf # Rust/Cargo
brew install mergiraf # macOS/Linux (if tap is available)
jq
JSON inspection used by various skills for structured output.
brew install jq # macOS/Linux
winget install jqlang.jq # Windows
apt-get install jq # Debian/Ubuntu
fd
Fast file discovery used as a fallback when tilth is unavailable.
brew install fd # macOS/Linux
cargo install fd-find # Rust/Cargo
winget install sharkdp.fd # Windows
apt-get install fd-find # Debian/Ubuntu
just
Project task runner used by /cook and /press to discover and run project commands.
brew install just # macOS/Linux
cargo install just # Rust/Cargo
winget install Casey.Just # Windows
Credits
The shared voice kernel at skills/age/references/voice.md β output discipline, reasoning posture, the certain | speculating | don't know confidence vocabulary, and the depth-vs-question split β adapts a Claude Opus 4.7 system-prompt experiment by Reebz, narrowed to the parts that earn their keep in a portable skills toolkit. Cross-referenced from briesearch, culture, mold, cook, and cure.
The /pasteurize skill β the six-phase diagnosis loop (feedback loop β reproduce β hypothesise β instrument β fix + regression test β cleanup) and the "build a feedback loop first" insight β adapts Matt Pocock's diagnose skill. Easy-cheese-specific adaptations (cheez-* tooling, handoff slug schema, --auto chain, /cook handoff for Phase 5) are layered on top.
The /wheypoint skill β compacting a conversation into a handoff document (with a suggested-skills section, no-duplication of existing artifacts, and secret redaction) β adapts Matt Pocock's handoff skill. Easy-cheese-specific adaptations: the handoff lands as a durable, resumable artifact at .cheese/notes/<slug>.md (rather than the OS temp directory) carrying the standard handoff slug, the suggested-skills section is a state-to-skill mapping over the cheese pipeline expressed as the slug's next: field plus named skills, and resumption runs through /cheese --continue <slug>.
The Superpowers brainstorming skill β adapted into /mold's design dialogue. Easy-cheese-specific adaptations: the agent-introduced-scope grep gate, shape-check blast-radius numbers, the two-key handshake (curdle gating on the user verb plus the agent's coherence self-check), and ADRs captured as a durable curdle by-product are ahead of the borrowed brainstorming flow.
The grill-with-docs skill (Pocock) β adapted into /mold's Grill phase with grounded, cited claims: every critical claim is verified via cheez-search / briesearch before seams are locked.
The cumulative-domain-model feature β the cumulative ubiquitous-language discipline that grows a durable domain-model.md domain model across sessions β adapts Matt Pocock's domain-modeling skill. Easy-cheese-specific adaptations layer the cheese pipeline's grounding and artifact conventions on top.
No comments yet
Be the first to share your take.