Configuration

tmux Theme

The theme lives in tmux/tmux.conf. It uses a dark blue-gray background with cyan accents.

Color Palette

Element	Color	Hex
Background	Dark blue-gray	`#1a1a2e`
Foreground	Warm gray	`#a8a8b0`
Accent (active borders, highlights)	Cyan	`#00d4ff`
Secondary	Blue	`#0066ff`
Tertiary	Purple	`#7c3aed`
Inactive borders	Muted indigo	`#333355`
Inactive elements	Zinc	`#71717a`
Active pane background	Lifted dark	`#1e1e34`

Visual Features (v2.1.0)

Role-Colored Pane Borders — active pane border color reflects the agent role:

Role	Color	Hex
Builder	Blue	`#0066ff`
Reviewer	Orange	`#f59e0b`
Tester	Yellow	`#eab308`
Optimizer	Purple	`#7c3aed`
Security	Red	`#ef4444`
Docs	Green	`#4ade80`

Pipeline Stage Badge — the status bar shows a live pipeline stage indicator (⚙ BUILD, ⚡ TEST, ↑ PR) with stage-specific colors.

Agent Count Widget — displays λN in the status bar showing the count of active agents from heartbeat files.

Active Pane Lift — subtle background depth effect distinguishes the active pane from inactive panes.

To customize colors, edit the hex values in tmux/tmux.conf and reload with prefix + r.

Keybindings

The prefix key is Ctrl-a (remapped from the default Ctrl-b).

General

Key	Action
`prefix + r`	Reload tmux config
`prefix + \|`	Split pane vertically
`prefix + -`	Split pane horizontally
`prefix + c`	New window
`prefix + x`	Kill pane (with confirmation)
`prefix + X`	Kill window (with confirmation)
`prefix + s`	Choose session (tree view)
`prefix + N`	New session

Key	Action
`prefix + h/j/k/l`	Move between panes
`Ctrl + h/j/k/l`	Smart pane switching (vim-tmux-navigator)
`prefix + H/J/K/L`	Resize pane (repeatable)

Agent Teams

Key	Action
`prefix + T`	Launch team session
`prefix + Ctrl-t`	Show team status dashboard
`prefix + g`	Display pane numbers
`prefix + G`	Toggle zoom on current pane
`prefix + S`	Toggle synchronized panes
`prefix + M-t`	Toggle team sync mode
`prefix + M-1`	Layout: main-horizontal (leader 65% left)
`prefix + M-2`	Layout: main-vertical (leader 60% top)
`prefix + M-3`	Layout: tiled (equal sizes)
`prefix + M-s`	Capture current pane scrollback
`prefix + M-a`	Capture all panes in window
`prefix + R`	Reap dead panes (one-shot cleanup)

Copy Mode (vi-style)

Key	Action
`v`	Begin selection
`y`	Copy selection
`r`	Toggle rectangle mode
`prefix + p`	Paste buffer

Claude Code Settings

The claude-code/settings.json.template configures Claude Code for agent teams.

Key Settings

Setting	Default	Description
`CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS`	`"1"`	Enable agent teams (required)
`CLAUDE_CODE_AUTOCOMPACT_PCT_OVERRIDE`	`"70"`	Context compaction threshold
`CLAUDE_CODE_SUBAGENT_MODEL`	`"haiku"`	Model for subagent lookups
`CLAUDE_CODE_MAX_TOOL_USE_CONCURRENCY`	`"5"`	Parallel tool calls per agent

Installation

Copy the template: cp claude-code/settings.json.template ~/.claude/settings.json
Customize the enabledPlugins section
Adjust environment variables as needed

Hooks

Hooks are shell scripts that run on Claude Code lifecycle events.

Included Hooks

Hook	Event	What It Does
`teammate-idle.sh`	TeammateIdle	Runs typecheck — blocks idle until errors are fixed
`task-completed.sh`	TaskCompleted	Runs lint + test — blocks completion until quality passes
`notify-idle.sh`	Notification	Desktop alert when agents need attention
`pre-compact-save.sh`	PreCompact	Saves git context before compaction

Installing Hooks

# Copy hooks to Claude Code hooks directory
cp claude-code/hooks/*.sh ~/.claude/hooks/
chmod +x ~/.claude/hooks/*.sh

Then wire them up in ~/.claude/settings.json:

{
  "hooks": {
    "teammate-idle": {
      "command": "~/.claude/hooks/teammate-idle.sh",
      "timeout": 30000
    }
  }
}

Writing Custom Hooks

Quality gate hooks use exit codes to control behavior:

Exit Code	Meaning
`0`	Pass — agent continues normally
`2`	Block — agent must fix the issue before proceeding
Other	Error — logged but doesn’t block

Example lint check hook:

#!/usr/bin/env bash
cd "$(find_project_root)" || exit 0
pnpm lint 2>&1 || {
  echo "::error::Lint errors found. Fix them before going idle."
  exit 2
}
exit 0

Daemon Intelligence Configuration

The daemon configuration file at .claude/daemon-config.json controls all daemon intelligence features. Generate it with shipwright daemon init.

Core Settings

Field	Default	Description
`watch_label`	`"ready-to-build"`	GitHub label to watch for
`poll_interval`	`60`	Seconds between polls
`max_parallel`	`2`	Maximum concurrent pipeline jobs
`pipeline_template`	`"autonomous"`	Default pipeline template
`base_branch`	`"main"`	Branch to create worktrees from
`model`	`"opus"`	Default Claude model
`skip_gates`	`true`	Auto-approve all gates

Adaptive Template Selection

Automatically selects the best pipeline template based on issue labels and triage score.

Field	Default	Description
`auto_template`	`false`	Enable adaptive template selection
`template_map`	`{}`	Regex-based template overrides (label pattern → template name)

When auto_template is enabled, template selection uses this priority:

Label overrides — hotfix/incident → hotfix template, security → enterprise
template_map overrides — regex patterns matched against issue labels
Score-based — score ≥ 70 → fast, ≥ 40 → standard, below → full

{
  "auto_template": true,
  "template_map": {
    "hotfix|incident": "hotfix",
    "security": "enterprise"
  }
}

Auto-Retry with Escalation

Automatically retries failed pipelines with optional model/template escalation.

Field	Default	Description
`max_retries`	`2`	Maximum retry attempts per issue
`retry_escalation`	`true`	Escalate model/template on retry

Self-Optimizing Metrics

Automatically tunes daemon parameters based on DORA metric thresholds.

Field	Default	Description
`self_optimize`	`false`	Enable self-optimizing loop
`optimize_interval`	`10`	Poll cycles between optimization checks

Self-optimization adjustments:

CFR > 40% — switches to full template
CFR > 20% — enables compound quality
Lead time > 4hrs — increases max_parallel, halves poll_interval
Lead time > 2hrs — enables auto_template
Deploy freq < 1/day — recommends merge stage
MTTR > 2hrs — recommends auto-rollback

Priority Lanes

Allows critical issues to bypass the normal queue and get processed immediately.

Field	Default	Description
`priority_lane`	`false`	Enable priority bypass queue
`priority_lane_labels`	`"hotfix,incident,p0,urgent"`	Labels that trigger priority lane
`priority_lane_max`	`1`	Maximum extra slots for priority issues

Org-Wide Mode

Watch issues across all repositories in a GitHub organization.

Field	Default	Description
`watch_mode`	`"repo"`	`"repo"` for single repo, `"org"` for org-wide
`org`	`null`	GitHub organization name (required for org mode)
`repo_filter`	`null`	Regex filter for repository names (e.g. `"api-.\|web-."`)

Patrol Settings

Proactive codebase scanning during quiet periods.

Field	Default	Description
`patrol.interval`	`3600`	Seconds between patrol runs
`patrol.max_issues`	`5`	Maximum issues to create per patrol
`patrol.label`	`"auto-patrol"`	Label applied to patrol-created issues

Degradation Alerting

Monitors pipeline success rates and alerts on quality declines.

Field	Default	Description
`alerts.degradation_window`	`5`	Number of recent pipelines to evaluate
`alerts.cfr_threshold`	`30`	Change failure rate (%) that triggers alert
`alerts.success_threshold`	`50`	Success rate (%) below which an alert fires

Health Monitoring

Field	Default	Description
`health.stale_timeout_s`	`1800`	Seconds before a job is considered stale

Lifecycle Hooks

Field	Default	Description
`on_success.remove_label`	`"ready-to-build"`	Label to remove on success
`on_success.add_label`	`"pipeline/complete"`	Label to add on success
`on_success.close_issue`	`false`	Close issue on success
`on_failure.add_label`	`"pipeline/failed"`	Label to add on failure
`on_failure.comment_log_lines`	`50`	Log lines to include in failure comment

Notifications

Field	Default	Description
`notifications.slack_webhook`	`null`	Slack webhook URL for alerts

Full Example

{
  "watch_label": "ready-to-build",
  "poll_interval": 60,
  "max_parallel": 2,
  "pipeline_template": "autonomous",
  "base_branch": "main",
  "model": "opus",
  "skip_gates": true,
  "auto_template": true,
  "template_map": {
    "hotfix|incident": "hotfix",
    "security": "enterprise"
  },
  "max_retries": 2,
  "retry_escalation": true,
  "self_optimize": true,
  "optimize_interval": 10,
  "priority_lane": true,
  "priority_lane_labels": "hotfix,incident,p0,urgent",
  "priority_lane_max": 1,
  "watch_mode": "repo",
  "org": null,
  "repo_filter": null,
  "patrol": {
    "interval": 3600,
    "max_issues": 5,
    "label": "auto-patrol"
  },
  "alerts": {
    "degradation_window": 5,
    "cfr_threshold": 30,
    "success_threshold": 50
  },
  "health": {
    "stale_timeout_s": 1800
  },
  "on_success": {
    "remove_label": "ready-to-build",
    "add_label": "pipeline/complete",
    "close_issue": false
  },
  "on_failure": {
    "add_label": "pipeline/failed",
    "comment_log_lines": 50
  },
  "notifications": {
    "slack_webhook": null
  }
}

Intelligence Configuration

The intelligence layer adds semantic codebase analysis, predictive risk assessment, and self-optimization to the daemon. All intelligence features are optional and configured in the intelligence section of .claude/daemon-config.json.

Core Intelligence Settings

Field	Default	Description
`intelligence.enabled`	`auto`	Master switch: `auto` enables when Claude CLI is available; `true`/`false` to override
`intelligence.cache_ttl_seconds`	`3600`	Cache duration for intelligence responses (prevents re-analyzing the same issue)

Intelligence Feature Flags

Field	Default	Description
`intelligence.composer_enabled`	`false`	Enable dynamic pipeline composition (replaces static template selection)
`intelligence.optimization_enabled`	`false`	Enable self-optimization (outcome analysis, template tuning, model A/B testing)
`intelligence.prediction_enabled`	`false`	Enable predictive analytics (risk assessment, anomaly detection, AI patrol)
`intelligence.adversarial_enabled`	`false`	Enable adversarial code review (red-team review after primary agent)
`intelligence.simulation_enabled`	`false`	Enable developer simulation (simulated PR review with multiple personas)
`intelligence.architecture_enabled`	`false`	Enable architecture enforcer (living architectural model and validation)

Intelligence Tuning Parameters

Field	Default	Description
`intelligence.ab_test_ratio`	`0.1`	Fraction of runs using composed pipelines for A/B testing
`intelligence.anomaly_threshold`	`3.0`	Standard deviations for anomaly detection

All intelligence features degrade gracefully when disabled or when Claude API calls fail. The system falls back to existing static behavior.

Auto-Scaling Configuration

The daemon supports resource-aware dynamic worker scaling based on system capacity and budget.

Field	Default	Description
`auto_scale`	`false`	Enable resource-aware dynamic worker scaling
`auto_scale_interval`	`5`	Seconds between scale checks
`max_workers`	`8`	Maximum workers the auto-scaler can allocate
`min_workers`	`1`	Minimum workers (floor)
`worker_mem_gb`	`4`	Memory per worker for capacity calculation
`estimated_cost_per_job_usd`	`5.0`	Cost estimate per job for budget-based scaling

Scaling factors (takes the minimum):

CPU: 75% of system cores
Memory: available GB / worker_mem_gb
Budget: remaining daily budget / estimated_cost_per_job_usd
Queue: current demand (active + queued issues)

Worker Pool Configuration

For fleet operations, distribute workers across multiple repositories.

Field	Default	Description
`worker_pool.enabled`	`false`	Enable shared worker pool across fleet repos
`worker_pool.total_workers`	`12`	Total workers to distribute across repos
`worker_pool.rebalance_interval_seconds`	`120`	Seconds between rebalance cycles

When enabled, the fleet rebalancer runs in the background and redistributes workers every N seconds. Repos with more queued issues get more workers. Each repo always gets at least 1 worker.

Fleet Configuration

The fleet configuration file at .claude/fleet-config.json manages multi-repo daemon orchestration. Generate it with shipwright fleet init.

{
  "repos": [
    { "path": "/path/to/api", "template": "autonomous", "max_parallel": 2 },
    { "path": "/path/to/web", "template": "standard" }
  ],
  "defaults": {
    "watch_label": "ready-to-build",
    "pipeline_template": "autonomous",
    "max_parallel": 2,
    "model": "opus"
  },
  "shared_events": true
}

Repo-Level Overrides

Each entry in the repos array can override the fleet defaults:

Field	Default	Description
`path`	—	Required. Absolute path to the repository
`template`	`defaults.pipeline_template`	Pipeline template override
`max_parallel`	`defaults.max_parallel`	Max concurrent pipelines for this repo
`watch_label`	`defaults.watch_label`	Label override for this repo
`model`	`defaults.model`	Model override for this repo

Fleet Defaults

Field	Default	Description
`defaults.watch_label`	`"ready-to-build"`	Default issue label to watch
`defaults.pipeline_template`	`"autonomous"`	Default pipeline template
`defaults.max_parallel`	`2`	Default max concurrent pipelines
`defaults.model`	`"opus"`	Default Claude model
`shared_events`	`true`	Share events.jsonl across repos

Deploy Configuration

Deploy configuration is set up via shipwright init --deploy or by using the deployed pipeline template. Deploy adapters live in scripts/adapters/.

Available deploy adapters:

Adapter	File	Description
Vercel	`adapters/vercel-deploy.sh`	Vercel deployment
Fly.io	`adapters/fly-deploy.sh`	Fly.io deployment
Railway	`adapters/railway-deploy.sh`	Railway deployment
Docker	`adapters/docker-deploy.sh`	Docker-based deployment

The deployed pipeline template uses placeholder commands (__STAGING_CMD__, __PRODUCTION_CMD__, __ROLLBACK_CMD__, __SMOKE_CMD__, __HEALTH_URL__) that are replaced during shipwright init --deploy.

Environment Variables

Variable	Default	Description
`CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS`	—	Required. Enables agent teams
`CLAUDE_CODE_SUBAGENT_MODEL`	parent model	Model for subagent lookups (`"haiku"` recommended)
`CLAUDE_CODE_AUTOCOMPACT_PCT_OVERRIDE`	`"80"`	Context compaction threshold (lower = more aggressive)
`CLAUDE_CODE_MAX_TOOL_USE_CONCURRENCY`	`"3"`	Parallel tool calls per agent
`SHIPWRIGHT_WEBHOOK_URL`	—	Custom webhook URL for pipeline notifications

tmux Plugins (TPM)

After installation, press prefix + I to install plugins.

Plugin	Key	What It Does
tmux-fingers	`prefix + F`	Vimium-style copy hints for URLs, paths, hashes
tmux-fzf-url	`prefix + u`	Fuzzy-find and open URLs
tmux-fzf	`F5`	Fuzzy finder for sessions, windows, panes
extrakto	`prefix + tab`	Extract and copy text from pane output
tmux-resurrect	auto	Save and restore sessions across restarts
tmux-continuum	auto	Automatic continuous session saving