Hosting the AI Slop Party

House Rules

The file set from The AI Slop Party. Rooms by blast radius, gates at the door, and agents that read the fridge.

Source on GitHub

Skills

Claude Code skills we wrote. Each one is a folder with a SKILL.md. Claude loads a skill when a request matches its description, or when you type /<skill-name>.

Contents

Skills in this set

Skill What it does Invoke
architecture-review Judges a PR, branch, path or proposal against the repo’s architecture of record: ARCHITECTURE.md, accepted ADRs, AGENTS.md + platform/ rules, and the architecture the code defines (module graph, dependency direction, layering, wire contracts). Runs as eng-manager-agent, which swarms principal-eng-agent, stacy-agent and the specialists the diff calls for, verifies every finding, and returns FOLLOWS / FOLLOWS WITH NOTES / ESCALATE with an escalation packet. Read-only. /architecture-review 123 (PR), a branch, a path, or “does this follow our architecture?”
smart-third-grader Rewrites a reply in the voice of a bright 8-year-old: short sentences, plain words, ideas kept correct. Also the house voice for code comments (see the comment-and-test-polish memory). “explain like a smart third grader”

skills/ is the source. .claude/skills/ is the identical copy Claude Code (and Copilot, Cursor) load in this repo. .agents/skills/ is generated by scripts/export-agents.py with spec-only frontmatter for Codex, Copilot, Cursor and Gemini CLI; outside Claude Code the Claude-only fields (context: fork, agent:, background:, argument-hint:) are dropped and the skill body says how to run without them.

How architecture-review runs

sequenceDiagram
  actor U as You or host-agent
  participant E as eng-manager-agent (forked)
  participant P as principal-eng-agent
  participant S as stacy-agent
  participant X as specialists (1–4)
  U->>E: /architecture-review SUBJECT
  E->>E: architecture of record @ origin/main SHA
  E->>E: scope: modules, stacks, rooms → pick team
  par in parallel
    E->>P: brief + lens: proven, simplest fit?
    E->>S: brief + lens: house rules, rooms, parity
    E->>X: brief + lens: stack conformance
  end
  P-->>E: cited findings
  S-->>E: cited findings
  X-->>E: cited findings
  E->>E: re-read every cited line, drop what fails
  E-->>U: verdict + findings + escalation packet

A finding counts only if it cites both the code (file:line) and the rule it is judged against. A rule is either written (ARCHITECTURE.md, an ADR, AGENTS.md) or “established” (the same pattern in two or more places on origin/main). With no written architecture, the report says so and labels every rule inferred.

Install

Target Command
personal (every project) cp -R skills/architecture-review skills/smart-third-grader ~/.claude/skills/
one repository cp -R skills/architecture-review skills/smart-third-grader ../your-app/.claude/skills/
a set of repositories install once personally, or repeat the repo copy per repo
Codex, Gemini CLI, or any Agent Skills runtime cp -R .agents ../your-app/ (or ~/.agents/skills/ for a personal install)

architecture-review needs the agents it calls installed in the same scope: eng-manager-agent (with the Agent tool, as shipped here), principal-eng-agent, stacy-agent and the specialists (see agents/README.md). It uses context: fork + agent: eng-manager-agent + background: false, so the review runs as eng-manager and you wait for the verdict.

Adding a skill

  1. Create skills/<name>/SKILL.md with name and description frontmatter. The description is the trigger, so list the phrases that should load it.
  2. Copy the folder to .claude/skills/<name>/ and add a row above.
  3. Keep the copies identical: diff -r skills/<name> .claude/skills/<name>, then scripts/export-agents.py to regenerate .agents/skills/<name>.