Agents Reference

All 18 agents — model, tools, memory scope, spawn triggers.

Agents are spawned by skills — they are not invoked directly. Each has memory scope for cross-session learning, a defined effort level that controls token budget, and — where relevant — is orchestrated via the new Agent Teams primitives (TeamCreate + SendMessage).

Agents at a glance

Agent	Model	Effort	Memory	Isolation	Tools
`comms-scanner`	Sonnet 4.6	low · 10	user	—	read-only
`infra-monitor`	Sonnet 4.6	low · 15	project	—	read-only + Vercel MCP
`project-scanner`	Sonnet 4.6	low · 15	project	—	read-only
`revenue-tracker`	Sonnet 4.6	medium · 20	project	—	read-only
`triage-agent`	Sonnet 4.6	high · 40	project	worktree	Edit/Write + Linear MCP
`daemon-agent`	Sonnet 4.6	low · 10	project	—	Bash
`doctor-agent`	Sonnet 4.6	high · 30	—	—	Edit/Write (no Agent spawn)
`memory-extractor`	Haiku 4.5	low · 10	project	—	read-only
`monitor-agent`	Haiku 4.5	low · 15	project	—	Bash, Read (read-only)
`yolo-ceo`	Opus 4.6	high · 20	user	—	strategic + Linear MCP
`yolo-cto`	Opus 4.6	high · 25	user	—	codebase analysis
`yolo-cfo`	Opus 4.6	high · 20	user	—	AWS + financial
`yolo-coo`	Opus 4.6	high · 25	user	—	execution + Linear MCP

Note

All worker agents upgraded from Sonnet 4.5 → Sonnet 4.6 in v0.8.0. See Changelog for rationale.

Background Scanner Agents

These run in the background, feeding structured JSON to the main skills.

`comms-scanner`

File: agents/comms-scanner.md
Model: claude-sonnet-4-6 · Effort: low · maxTurns: 10
Memory: user scope · Read-only: Write/Edit disallowed
Tools: Bash, Read
Purpose: Scans all communication channels (WhatsApp, Email, Slack, Telegram) for full inbox state. Classifies each conversation as NEEDS_REPLY / WAITING / HANDLED. Returns structured JSON consumed by ops-inbox and ops-go.
Spawned by: /ops:inbox (as part of TeamCreate("inbox-channels")), /ops:go
Initial prompt: "Scan all configured channels and classify conversations. Return structured JSON."

`infra-monitor`

File: agents/infra-monitor.md
Model: claude-sonnet-4-6 · Effort: low · maxTurns: 15
Memory: project scope · Read-only: Write/Edit disallowed
Tools: Bash, Read, Vercel MCP (list_projects, list_deployments, get_deployment, get_runtime_logs)
Purpose: ECS, Vercel, and AWS health checker. Returns structured JSON with service health, recent deployments, and anomaly flags.
Spawned by: /ops:fires, /ops:deploy

`project-scanner`

File: agents/project-scanner.md
Model: claude-sonnet-4-6 · Effort: low · maxTurns: 15
Memory: project scope · Read-only: Write/Edit/Agent disallowed
Tools: Bash, Read, Glob
Purpose: Git, PR, and CI status scanner across all registered repos. Returns structured JSON with branch state, uncommitted files, open PRs, and CI status.
Spawned by: /ops:projects, /ops:go

`revenue-tracker`

File: agents/revenue-tracker.md
Model: claude-sonnet-4-6 · Effort: medium · maxTurns: 20
Memory: project scope · Read-only: Write/Edit/Agent disallowed
Tools: Bash, Read
Purpose: Revenue, billing, and credits analysis. AWS Cost Explorer, credits balance, project revenue stages, burn rate, runway.
Spawned by: /ops:revenue

Fix Agents

`triage-agent`

File: agents/triage-agent.md
Model: claude-sonnet-4-6 · Effort: high · maxTurns: 40
Memory: project scope
Isolation: 🏝️ worktree (sandboxed)
Tools: Bash, Read, Grep, Glob, Edit, Write, Linear MCP (get_issue, update_issue, save_comment)
Purpose: Investigates a specific issue from Sentry, Linear, or GitHub. Finds root cause, checks if already fixed, creates fix branch + PR if needed.
Input: ISSUE_SOURCE, ISSUE_ID, ISSUE_TITLE, ISSUE_BODY, AFFECTED_REPO, AFFECTED_FILE (optional)
Spawned by: /ops:triage (via TeamCreate("triage-fixers")), /ops:fires

Tip

The worktree isolation means each triage agent runs in an isolated copy of the repo — changes don't pollute the main working tree until a PR is opened.

Daemon & System Agents

`daemon-agent`

File: agents/daemon-agent.md
Model: claude-sonnet-4-6 · Effort: low · maxTurns: 10
Memory: project scope
Purpose: Manages the ops-daemon — start, stop, restart, reconcile services, check health. Spawned by /ops:doctor and /ops:setup when daemon configuration changes are needed.

`doctor-agent`

File: agents/doctor-agent.md
Model: claude-sonnet-4-6 · Effort: high · maxTurns: 30
Tools: Bash, Read, Write, Edit, Grep, Glob (no Agent spawning)
Purpose: Diagnoses and auto-fixes plugin configuration errors, manifest issues, broken permissions, invalid JSON, stale cache copies. Calls bin/ops-autofix for known-fix categories (wacli FTS5 rebuild, Slack MCP registration, Vercel MCP registration).
Spawned by: /ops:doctor

`memory-extractor`

File: agents/memory-extractor.md
Model: claude-haiku-4-5-20251001 · Effort: low · maxTurns: 10
Memory: project scope
Purpose: Extracts user profiles, contact cards, and behavioral patterns from chat history. Runs as a daemon service every 30 minutes. Writes structured markdown to ~/.claude/plugins/data/ops-ops-marketplace/memories/.
Used by: all communication skills for context-aware drafting
Cost: <$0.01 per run (Haiku 4.5)

See Memories System.

Monitoring Agents

`monitor-agent`

File: agents/monitor-agent.md
Model: claude-haiku-4-5 · Effort: low · maxTurns: 15
Memory: project scope · Read-only: Write/Edit/Agent disallowed
Tools: Bash, Read
Purpose: Lightweight APM and metrics probe. Queries Datadog (monitors + event stream), New Relic (NerdGraph alerts + entity health), and OpenTelemetry (OTLP/HTTP endpoints) for active alerts, error traces, and entity health. Returns structured JSON consumed by /ops:monitor.
Spawned by: /ops:monitor (including --watch live polling every 60s)

Important

monitor-agent runs on Haiku 4.5, not Sonnet 4.6. APM polling is high-frequency (every 60s in watch mode) and latency-sensitive. Haiku 4.5 delivers the same structured-JSON quality at ~10× lower cost and ~2× lower latency.

Cost: <$0.001 per poll · --watch runs ~$0.05/hour

C-Suite Analysis Agents

All four run in parallel when /ops:yolo is invoked, via TeamCreate("yolo-csuite"). Each uses Opus 4.6 for maximum analytical depth.

flowchart LR
    Y["/ops:yolo"] --> G[Pre-gather data]
    G --> T[TeamCreate<br/>yolo-csuite]
    T --> CEO[yolo-ceo]
    T --> CTO[yolo-cto]
    T --> CFO[yolo-cfo]
    T --> COO[yolo-coo]
    CEO -.->|writes| F0[/tmp/yolo-session/ceo-analysis.md]
    CTO -.->|writes| F1[/tmp/yolo-session/cto-analysis.md]
    CFO -.->|writes| F2[/tmp/yolo-session/cfo-analysis.md]
    COO -.->|writes| F3[/tmp/yolo-session/coo-analysis.md]
    F0 & F1 & F2 & F3 --> Main[Main /ops:yolo<br/>orchestrator<br/>synthesizes all 4]
    Main --> R[Hard Truths Report]

    classDef opus fill:#8b5cf6,color:#fff,stroke:#6d28d9
    classDef primary fill:#6366f1,color:#fff,stroke:#4338ca
    class CEO,CTO,CFO,COO opus
    class Main,R primary

Important

Who synthesizes? The main /ops:yolo skill orchestrator (running in Claude Code) reads all 4 agents' analysis files and synthesizes them into the Hard Truths report. yolo-ceo is NOT the synthesizer — it's a peer that produces its own strategic perspective in parallel with the other three.

`yolo-ceo`

File: agents/yolo-ceo.md
Model: claude-opus-4-6 · Effort: high · maxTurns: 20
Purpose: Strategic CEO perspective — runs in parallel with CTO/CFO/COO, writes its own analysis. Does NOT read or synthesize the other agents' files.
Tools: Bash, Read, Write, Grep, Glob, Linear MCP (list_issues, list_projects)
Covers: Growth blockers (#1 thing blocking growth) · Resource allocation · Build vs. buy · Biggest time sinks · Honest investor pitch · What to start over on this month

`yolo-cto`

File: agents/yolo-cto.md
Model: claude-opus-4-6 · Effort: high · maxTurns: 25
Tools: Bash, Read, Write, Grep, Glob
Covers: Worst tech debt · Which service is one bad deploy from a P0 · What breaks at 10× load · Corners cut

`yolo-cfo`

File: agents/yolo-cfo.md
Model: claude-opus-4-6 · Effort: high · maxTurns: 20
Tools: Bash, Read, Write
Covers: Actual vs. perceived burn rate · Waste (orphaned resources) · Runway · ROI · Credits expiry

`yolo-coo`

File: agents/yolo-coo.md
Model: claude-opus-4-6 · Effort: high · maxTurns: 25
Tools: Bash, Read, Write, Grep, Glob, Linear MCP (list_issues, list_cycles)
Covers: Stale PRs (>7 days) · Stalled Linear issues (5+ days no activity) · Stuck GSD phases · Abandoned repos (no commits in 2+ weeks) · Communication failures

Caution

Rule 5 applies to C-suite recommendations. Any destructive infrastructure change (delete ECS cluster, stop service, cancel domain, rewrite git history) is flagged with ⚠️ REQUIRES CONFIRMATION in the output — the orchestrator (or user) must explicitly approve before execution.

Agent memory scopes

Scope	What persists
`project`	Remembered across sessions within the same project context
`user`	Remembered across all projects for the same user
worktree isolation	Agent runs in a sandboxed worktree; changes don't bleed into main tree

Agent Teams (new in v0.8.0)

Skills that dispatch multiple parallel agents use the TeamCreate + SendMessage primitives:

Skill	Team name	Agents
`/ops:fires`	`fire-fixers`	one agent per incident
`/ops:inbox`	`inbox-channels`	one per channel (WhatsApp/Email/Slack/Telegram)
`/ops:merge`	`merge-fixers`	one per repo
`/ops:orchestrate`	`wave-N-cross-repo`	multiple agents coordinating via `SendMessage`
`/ops:triage`	`triage-fixers`	one per active issue
`/ops:yolo`	`yolo-csuite`	CEO/CTO/CFO/COO in parallel

SendMessage is used for cross-agent coordination — e.g., when api-worker finishes a schema change, it sends a message to mobile-worker with the new types so the downstream agent can proceed without re-dispatching.

v2.0 Auto-Fix Agents

These agents are dispatched headlessly by the Auto-Fix Subsystem. They are never invoked directly.

`build-fixer`

File: agents/build-fixer.md
Model: claude-haiku-4-5 · Effort: high · maxTurns: 30
Memory: none · Isolation: none
Tools: Read, Edit, Bash, Grep, Glob
Purpose: Repairs a single failed local build (npm run build:*, fastlane, expo, etc.). Diagnoses TypeScript errors, bundler failures, SDK version drift, and Pod compilation issues. Opens a fix PR and exits.
Dispatched by: bin/ops-deploy-fix-build-trigger (PostToolUse:Bash hook)

`deploy-fixer`

File: agents/deploy-fixer.md
Model: claude-haiku-4-5 · Effort: high · maxTurns: 30
Memory: none · Isolation: none
Tools: Read, Edit, Bash, Grep, Glob
Purpose: Diagnoses and remediates a single failed post-merge deployment. Checks GitHub Actions logs, ECS health, Vercel errors. Opens one fix PR and exits.
Dispatched by: scripts/ops-deploy-monitor.sh (background watcher after gh pr merge)

`dependency-auditor`

File: agents/dependency-auditor.md
Model: claude-haiku-4-5 · Effort: medium · maxTurns: 20
Memory: none · Read-only: does not modify lockfiles
Tools: Read, Bash, Grep, Glob, WebFetch
Purpose: Audits project dependencies for security advisories, version drift, and unused packages. Surfaces findings and a recommended action plan. Does NOT auto-upgrade major versions.
Spawned by: Manual invocation or future scheduled cron.

`general-purpose`

File: agents/general-purpose.md
Model: claude-sonnet-4-6 · Effort: high · maxTurns: 40
Memory: none
Tools: Read, Write, Edit, Bash, Grep, Glob, NotebookEdit, TodoWrite, WebSearch, WebFetch
Purpose: Fully-equipped fallback agent for tasks that don't fit a known specialist. Knows when to delegate to specialists vs do the work itself. Used when no other claude-ops agent matches AND the task spans multiple domains.
Spawned by: The specialized-agent auto-suggestion hook when no keyword match is found.

Agents Reference

Agents Reference

Agents at a glance

Background Scanner Agents

comms-scanner

infra-monitor

project-scanner

revenue-tracker

Fix Agents

triage-agent

Daemon & System Agents

daemon-agent

doctor-agent

memory-extractor

Monitoring Agents

monitor-agent

C-Suite Analysis Agents

yolo-ceo

yolo-cto

yolo-cfo

yolo-coo

Agent memory scopes

Agent Teams (new in v0.8.0)

v2.0 Auto-Fix Agents

build-fixer

deploy-fixer

dependency-auditor

general-purpose

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

claude-ops Wiki

Getting started

Reference

v2.0 — autonomy layer

v2.4 — smart home

v2.3 — competitor intel

v2.11 — telegram bot push · rotation improvements

v2.9 — voice & phone

Usage

Internals

Clone this wiki locally

`comms-scanner`

`infra-monitor`

`project-scanner`

`revenue-tracker`

`triage-agent`

`daemon-agent`

`doctor-agent`

`memory-extractor`

`monitor-agent`

`yolo-ceo`

`yolo-cto`

`yolo-cfo`

`yolo-coo`

`build-fixer`

`deploy-fixer`

`dependency-auditor`

`general-purpose`