Shared agents that monitor, maintain, and improve every project in the fleet. The map below shows how they connect — click any agent for its full detail.
How it works
The Infinite Work Loop
You feed in todos; the pack finds, routes, executes, and verifies the work — then loops back for more. It never stops.
Click any agent in the map for its mission, schedule, and status.
Understanding the Pack
A field guide to how the wolfpack is built — what each kind of agent does, and how the work flows. The map above shows it; this explains it.
Pack
Pack — the one coordinator
Pack is the fleet's brain. It doesn't do the work itself — it takes the todos you feed in, routes each to the right agent, sets the standards every project follows, and keeps the loop turning. There is exactly one Pack, and every other agent works beneath it. You — the human — sit above Pack and point it where to go.
Every other agent answers two questions:
SCOPEwhat it watches — the whole fleet, one project, or one package.
KINDwhat it does — hub, specialist, service, or wolf.
Hubs
fleet
Broad sensors that find the work and fan it out to everyone else.
Builders — each owns one app: ships features, fixes bugs, runs the roadmap. They live inside their project.
Nova · Seeker · one per app
The loop, in words: you feed in todos → hubs & services find and route the work → Pack assigns it → a wolf executes on a branch → a service verifies the result → it loops back for more. It never stops.
Flows & Use Cases
Concrete plays the pack runs on repeat — each turns a slow, manual chore into an automated loop. Here's the path, and what it buys you.
Ship a todo, hands-off
idea → PR in hours
Drop a todo and walk away — it comes back as a reviewable PR.
Youtodo→Clerkclassify + route→Wolfbuild on a branch→Youreview the PR→Shipmerge + deploy
Before: you triage it, switch context, and write the change yourself. After: one todo in, a PR out — you only review and merge.
Same-day security patch
advisory → patched same day
A new CVE doesn't sit in a backlog for weeks across your repos.
Before: dependency debt piles up silently until a build breaks. After: weekly bumps, tested and shipped without you touching them.
How the Pack Runs
Three ways the work actually executes. Most agents are scheduled scripts; Scout & Counsel are also invokable subagents; skills are lanes you trigger while building.
⏰
Scheduled script
Most agents are a bash script in wolf/scripts/ run on a timer by launchd (macOS) or cron. They wake, scan their domain, and file todos — no human in the loop.
# run once, by hand
bash wolf/scripts/repo-auditor.sh
# or schedule it (macOS launchd)
cp schedules/com.latentfreedom.counsel-fleet.plist \
~/Library/LaunchAgents/
launchctl load \
~/Library/LaunchAgents/com.latentfreedom.counsel-fleet.plist
🤖
Claude Code subagent
Scout and Counsel also exist as invokable subagents — delegate a task in a Claude Code session and they run with their own prompt + tools.
# symlink the definition once
ln -s /Users/nick/Sites/wolf/claude-agents/scout.md \
~/.claude/agents/scout.md
# then, in a session:
"use the scout subagent to scan for advisories"
⚡
Skill / lane
Skills are workflows you invoke while working — not scheduled. Trigger a lane and Claude follows its playbook: read context, branch, validate, open a PR.
# invoke the lane that drives this dashboard
/live-project-roadmap-orchestrator
Inside a Wolf Run
The full picture — the harness a wolf runs in, the loop it thinks in, the memory it draws on, and the fleet ops feedback loop that makes every next run better. Our wolf-named take on the standard agent runtime.
Reading it: a todo + the wolf's SOUL.md and memory load into context → the wolf loops with its tools until guardrails pass → it ships a PR → Fleet Ops traces, evaluates, and (if it passed) releases improvements back into the wolf's SOUL + standards. It never stops.