Agents

Agents (brewers) are specialized roles defined in markdown files with frontmatter metadata. Each agent has its own model, toolset, and system prompt. Brewcode ships 5 agents (4 public + 1 internal helper) — specialists only, for building skills, agents, hooks, and shell scripts.

Generic roles (developer, tester, reviewer, architect) are no longer shipped: every project generates its own agents into .claude/agents/ via /brewcode:teams-setup or /brewcode:agents, tuned to that project’s stack. Skills that need a reviewer or an implementer pick the project’s agent from .claude/agents/, falling back to general-purpose.

Summary table

Purpose column = the agent’s own description frontmatter, verbatim. That string is what Claude matches against when auto-selecting an agent.

#AgentModelWrite toolsDescription (verbatim)
1skill-creatorinheritRead, Write, Edit, BashCreates and improves Claude Code skills. Triggers: create skill, improve skill, fix activation.
2agent-creatorinheritRead, Write, Edit, BashCreates and improves Claude Code agents. Triggers: create agent, improve agent, scaffold agent.
3hook-creatorinheritRead, Write, Edit, BashCreates and debugs Claude Code hooks. Triggers: create hook, PreToolUse hook, debug hook.
4bash-expertinheritRead, Write, Edit, BashCreates sh/bash scripts for Mac/Linux. Triggers: create script, bash script, shell script.
5bc-rules-organizerhaikuRead, Write, Edit, BashInternal. Spawned only by /brewcode:rules. No direct/auto use.

model: inherit means the agent runs on whatever model the session runs on — no per-agent pin. Only the internal bc-rules-organizer (haiku) is pinned.

🧩

skill-creator

Skill creation and improvement

👫

agent-creator

Agent creation and improvement

🔗

hook-creator

Hook creation and debugging

📟

bash-expert

Bash/sh script creation

🔗

bc-rules-organizer

Internal — rules file organization

Internal agent

bc-rules-organizer is internal — spawned only by /brewcode:rules, never user-invokable directly.

Creator contract suite

agents/tests/suite-creator-contract.mjs is a 27-check regression suite that pins what hook-creator, agent-creator and skill-creator teach about the Claude Code hook and subagent API, so the next upstream change fails a test instead of silently teaching a stale one.

Scope guard — shared by all 4 public agents

Every public brewcode agent sizes the task before starting. One bounded unit = one deliverable, roughly 5 files, roughly 10 steps. Anything larger is refused up front.

SituationWhat the agent does
Brief fits one bounded unitStarts immediately
Brief exceeds ~5 files / ~10 steps, or bundles several independent deliverablesStops before starting, returns a split proposal: 2-N bounded subtasks, each with scope and a suggested owner
Scope grows mid-flightStops at the next clean boundary, reports done / remaining / how to split
Brief is missing GOAL, SCOPE, CONTEXT, CONSUMER or acceptanceStates the assumption explicitly in the report, or asks once — never invents scope

Agents deliver for the CONSUMER, not the literal wording: the result must be usable as-is by whoever picks it up next, with the whole briefed scope covered.

Got a split proposal instead of a result?

The brief was too big — this is the guard working, not a failure. Pick one subtask from the proposal and re-send it, or spawn the subtasks as parallel agents in a single message.

Models

ModelAgentsMeaning
inheritskill-creator, agent-creator, hook-creator, bash-expertRuns on the session’s model — you control the tier
haikubc-rules-organizerPinned: cheap file reorganization

Automatic selection

Agents are launched automatically by brewcode skills via the Task API. Each skill spawns the relevant agent for its domain — no manual agent selection required.

Moved Agents

Agents text-optimizer, ssh-admin, and deploy-admin have been moved to Brewtools.

🚀

Latest Release

Download, changelog, and installation instructions.

🔗

View on GitHub

Source code, README, and configuration files.