Full Setup

11 setups 27 skills · 8 agents user-invoked

Run /brewcode:setup-status first. Its run-list puts search and conventions before the agents that consume them, then installs task management and finishes with documentation sync. Run each setup yourself, preferably in a fresh session, and inspect the generated files before committing them.

A -setup skill installs or maintains agents, hooks, generated skills, rules, or a task board. Use the installed mechanism afterwards; rerun its setup to inspect, refresh, disable, or remove it. Tools such as /brewtools:text-optimize and /brewdoc:publish remain recurring commands.

Start here: the dashboard

/brewcode:setup-status
/brewcode:setup-status brewtools
/brewcode:setup-status convention-setup

The first command reports all installed plugins; the others filter by plugin or setup. Rows show found artifacts, content versions, producing plugin versions, and the next command. The dashboard does not run setups and has no --run mode. Its one optional write is an explicitly accepted task-tools settings key; the setup probes themselves are read-only.

Recommended order

This follows the seven stages in setup-status. Its run-list omits installed, deliberately disabled, unavailable, and filtered-out rows. Within each stage, partial installs precede stale and missing ones.

What the eleven setups install

Skill	Plugin	Main artifacts
agent-return-setup	brewtools	Three shared hook files and .claude/agent-return.json; SubagentStart contract and SubagentStop return guard
semble-setup	brewcode	User-scope semble_code MCP server, five project hook files wired to six events, .claude/rules/semble-first.md, .claude/semble/, .sembleignore, and a managed CLAUDE.md block
convention-setup	brewcode	.claude/convention/{reference-patterns,testing-conventions,project-architecture}.md and .claude/rules/convention.md
teams-setup	brewcode	Team manifests, trace files, project agents, and shared intent-guard
agent-router-setup	brewtools	.claude/hooks/agent-router.mjs and .claude/brewtools/agent-router.json
manager-setup	brewtools	Manager hard-wall guard, state and disarm helper under .claude/brewtools/manager/, plus settings registration
agent-deadline-setup	brewtools	Deadline guard and cleanup hooks, with .claude/agent-deadline.json
superreview-setup	brewcode	Project /superreview skill, references, pristine template baseline, and shared intent-guard
task-board-setup	brewtools	Board tree, task-tracker, task-board skill, tasks rule, PROGRESS.md, METHODOLOGY.md, ANTI-DRIFT.md, and task-graph.md; task-spec skill and spec/design templates only when you accept SPEC_MODE=on
memory-sync-setup	brewdoc	Project /memory-sync skill and memory-guide, agent-audit, and hard-sync references
docsync-setup	brewdoc	Three doc-staleness hooks and .claude/docsync/ config and state

Agent-return and agent-deadline also accept global scope under ~/.claude/. Semble registers its MCP server at user scope. Other listed artifacts belong to the target project. Manager’s codeword prompt hook ships with the plugin; the hard wall is installed separately.

Task methodology and session anti-drift

Task-board setup generates shared domain guidance at .claude/features/METHODOLOGY.md. Every accepted task gets its own Methodology and Anti-drift cron sections before implementation. They record the goal, review and test strategy, base work units, and a complete task-specific tick prompt.

When a top-level task becomes active, the main session’s task-board flow reconciles one session timer for it. The default is hourly (0 * * * *); your requested cadence or opt-out takes precedence. Child work units use their parent’s timer. Task-tracker reconciles task records and returns timer actions to the main session.

Each delivered tick:

• Rereads shared and task methodology, anti-drift instructions, goal, spec, user corrections, and task records.
• Requests concise evidence from available active agents and reconciles statuses, dependencies, counts, and next work.
• Rebuilds task-graph.md, keeping every unfinished node and the latest ten completed nodes; preserves older completion evidence in task records before pruning graph rows.
• Checks drift against the goal and acceptance evidence, corrects deviations within scope, and advances the next unblocked authorized step.
• Reports local time and timezone, tick number, elapsed time, achievements, remaining work, next action, and any blockers or user questions in at most five short lines using 🟢 🔵 🔴 ⚪.

Task-board announces the confirmed timer id, cadence, timezone, known due time, session scope, and stop/change controls. It verifies stopping on completion, cancellation, or parking. If scheduling tools are unavailable, it saves the prompt and reports that scheduling is unavailable. Stored task state does not keep a timer alive after a session ends; scheduled time and actual receipt can differ.

Typing +++ in Plan mode adds an instruction to include methodology, graph reconciliation, the task-specific prompt, and a timer step in the execution plan. Planning remains read-only: files and timers change during authorized execution. Outside Plan mode, +++ adds no cron instruction.

Read the setup verdicts

Verdict	Meaning
n/a	The owning plugin is not installed
disabled	A live off-switch or parked entry file shows deliberate disablement
missing	No anchor or secondary artifact is present
partial	Incomplete artifacts, half-applied toggles, unresolved metadata, or incorrect ownership
stale	Artifact content version, compared bytes, or a required artifact/wiring signal needs refresh
installed	Required artifacts, content-version signals, and applicable byte comparisons agree

content_version is the headline freshness signal; version identifies the plugin release that produced the installation. A newer plugin release alone does not make unchanged artifact content stale. Generated project content has no meaningful byte-for-byte comparison with a raw template. The dashboard also checks generated_by ownership and copied assets where applicable.

Disabled state takes precedence over missing, partial, and stale. Config-based setups switch a live flag; teams, superreview, task-board, conventions, and memory-sync park discovery entry files. Disabling conventions parks the loading rule only: accepted coding rules and manual CLAUDE.md references remain active. A missing plugin comparison asset is a reported validation gap, not evidence that the installed file is stale.

Shared modes

Mode	Purpose
status	Inspect without changing artifacts
install	Generate or copy artifacts and wire the mechanism
upgrade	Refresh from current source while preserving supported local configuration and disabled state
enable / disable	Resume or pause the installed mechanism
uninstall	Remove installation artifacts or wiring according to that setup’s ownership boundary
purge	Remove the additional owned data specified by that setup

All eleven accept these seven modes. Use explicit modes: many setups choose status when installed and install when absent, but conventions default to install and semble always defaults to status. Convention setup also accepts full, conventions, rules, and paths; semble adds reindex, optimize, and resume. Each skill page documents exact scope, arguments, and removal boundaries.

Disarm the manager wall

The installed manager hard wall can block main-session file and shell tools. Its narrow helper exemption lets you disarm it:

/brewtools:manager-setup disable

The skill runs the installed helper in this form:

node /abs/path/to/project/.claude/brewtools/manager/manager-state.mjs set hard=false

Use the actual project path and helper CLI directly. Shell operators, expansions, and node eval flags invalidate the exemption. If an old installation lacks the helper, have a subagent perform manager-setup upgrade, then disable the wall through the installed helper.

After setup

Rerun /brewcode:setup-status after plugin updates. Follow its scope-specific commands to repair partial installs or refresh changed artifacts. Plugin installation alone does not replace files that a setup previously copied into your project.

Related