Solo Squad

One developer with AI ships like a team of twenty.

27 skills + 6 workflow commands that turn AI agents into disciplined engineers. Each skill is a battle-tested process with DOT flowcharts, hard gates, and anti-slop defenses — so your agent follows the playbook instead of improvising.

Zero dependencies. Pure markdown + shell scripts. Install in 10 seconds.

Install

No dependencies. No build step. Pure markdown + shell scripts.

Claude Code (Recommended)

Marketplace Install (Recommended)

claude plugins marketplace add https://github.com/efecanbasoz/solo-squad
claude plugins install solo-squad

Reload to activate:

/reload-plugins

Manual Install

git clone https://github.com/efecanbasoz/solo-squad ~/.claude/solo-squad

Then add to your Claude Code plugins directory. Skills are auto-discovered from .claude-plugin/plugin.json.

Codex CLI

Marketplace Install (Recommended)

codex plugin marketplace add efecanbasoz/solo-squad
codex plugins install solo-squad

Or add the marketplace manually:

# Clone the repo
git clone https://github.com/efecanbasoz/solo-squad ~/.codex/solo-squad

# Add the marketplace
cp ~/.codex/solo-squad/.agents/plugins/marketplace.json ~/.agents/plugins/solo-squad-marketplace.json

Restart Codex to discover the plugin.

Manual Install (Legacy)

git clone https://github.com/efecanbasoz/solo-squad ~/.codex/solo-squad
mkdir -p ~/.agents/skills
ln -s ~/.codex/solo-squad/skills ~/.agents/skills/solo-squad

To enable subagent dispatch for skills like /build and /sprint, add to ~/.codex/config.toml:

[features]
multi_agent = true

OpenCode

Plugin Registry Install (Recommended)

opencode plugin "solo-squad@git+https://github.com/efecanbasoz/solo-squad.git" --global

Or add it manually to your opencode.json:

{
  "plugin": ["solo-squad@git+https://github.com/efecanbasoz/solo-squad.git"]
}

Manual Install

git clone https://github.com/efecanbasoz/solo-squad ~/.config/opencode/solo-squad

Then add to opencode.json:

{
  "plugin": ["/absolute/path/to/solo-squad"]
}

The JS plugin at .opencode/plugins/solo-squad.js registers skills, workflow commands, and subagents automatically, then injects session context on the first user message.

Updating

Each platform picks up updates differently:

Platform	Update command
Claude Code	`claude plugins update solo-squad`
Codex CLI	`codex plugin marketplace upgrade` or `cd ~/.codex/solo-squad && git pull`
OpenCode	Remove + re-add in `opencode.json` (auto-fetches latest)

Uninstalling

Platform	Uninstall command
Claude Code	`claude plugins uninstall solo-squad`
Codex CLI	`codex plugins uninstall solo-squad`
OpenCode	Remove from `opencode.json`

What makes this different

Most AI coding plugins give you vague prompts. Solo Squad gives you processes agents can't skip.

Other plugins	Solo Squad
Vague instructions	DOT flowcharts — agents follow structured diagrams, not prose
Agents skip steps	Hard gates (`<HARD-GATE>`, `<SUBAGENT-STOP>`) — behavioral boundaries that can't be rationalized away
"Looks good to me" reviews	Confidence calibration — every finding rated 100/75/50/25. Only 8/10+ gets reported. Zero noise.
Generic code review	Evidence-based certification — Reality Checker defaults to `NEEDS WORK`. Overwhelming proof required for `READY`.
No memory between sessions	Persistent learnings — `/compound` and `/learn` capture patterns to `~/.solo-squad/learnings.jsonl` so your agent gets smarter over time
Hidden dependencies	Zero dependencies — pure markdown + shell scripts. No npm install, no compiled binaries, no Docker.
"AI slop" in output	Anti-slop scan — `/slop-scan` catches empty catches, redundant awaits, dead code, and copy-paste smell before review

The result: Your agent actually follows the process. No more skipped tests, no more "looks good to me" on broken code, no more hallucinated deployments.

See it work

Ship a feature in one command

You: /sprint

Solo Squad: Starting sprint. Phase 1: /brainstorm
  What are you building and why?

You: Real-time notifications for the dashboard. Users miss critical alerts.

Solo Squad: Before we code — three questions:
  1. WebSocket, SSE, or polling? (trade-offs below)
  2. What's the simplest version that delivers value?
  3. What are you intentionally NOT building?

[5 min conversation — design doc produced]

Solo Squad: Design doc saved. → /plan

  12 tasks created (2-5 min each):
  ├── #1  NotificationProvider context
  ├── #2  WebSocket hook with reconnect
  ├── #3  Toast component with severity levels
  ├── ...
  └── #12 E2E test: notification lifecycle

  Continue to /autoplan? [y/N]

You: y

Solo Squad:
  === AUTOPLAN VERDICT ===
  Overall: PROCEED

  CEO:    CLEAR — Scope matches the user problem
  Design: PASS  — Main UI flow is covered
  Eng:    CLEAR — Data flow and reconnect edge cases are explicit
  DevEx:  PASS  — Local verification path is straightforward

  → /build

[Subagents execute each task with TDD — 25 min]

Solo Squad: Build complete. 12/12 tasks, 94% coverage.
  → /review

  Multi-lens review:
    🔴 BLOCKER: Race condition in WebSocket reconnect (auto-fixed)
    🟡 SUGGESTION: Extract toast queue into custom hook
    💭 NIT: Inconsistent error message casing

  → /polish-beta

  ⏸ POLISH #1 — docs/notifications.md:42
  Current:  "Alerts are send instantly"
  Proposed: "Alerts are sent instantly"
  Reply: apply

You: apply

Solo Squad:
  ✓ Polish applied
  → /ship

  ✓ Tests passing (47/47)
  ✓ Coverage: 94% (threshold: 80%)
  ✓ PR opened: feat(notifications): add real-time alert system

[Review approved]

Solo Squad:
  → /land-and-deploy

  ✓ Merge complete
  ✓ Deploy ready: https://example.vercel.app
  ✓ Canary: PROCEED
  ✓ Docs synced: CHANGELOG.md, README.md
  → /compound

Catch what CI misses

You: /security-scan

Solo Squad:
  CRITICAL  SQL injection in /api/users?search= — parameterize query
  HIGH      No rate limiting on /api/auth/login — add express-rate-limit
  MEDIUM    jsonwebtoken@8.5.1 has CVE-2022-23529 — upgrade to 9.x

  3 findings (8/10+ confidence). Auto-fix available for 2. Proceed? [y/N]

Get a design review that actually helps

You: /design-review

Solo Squad:
  Visual Quality:    7/10  — spacing inconsistent between card variants
  Consistency:       5/10  — three different border-radius values used
  Accessibility:     6/10  — contrast ratio 3.2:1 on muted text (need 4.5:1)
  Brand Alignment:   4/10  — off-brand colors in new components
  AI Slop Detection: CLEAN — no generic stock patterns detected

  Top fix: Unify border-radius to design token --radius-md
  Before/after diff generated.

What's inside

Development — 16 skills, 5 agents

Skill	Your specialist
`/brainstorm`	💡 Product partner who challenges your assumptions
`/plan`	🗺️ Architect who breaks work into 2-5 min tasks with file paths
`/plan-review`	🔍 Multi-lens review: CEO, Design, Eng, DevEx in one pass
`/build`	🔨 Lead engineer dispatching subagents with TDD enforcement
`/review`	👁️ Senior reviewer classifying 🔴 BLOCKER / 🟡 SUGGESTION / 💭 NIT
`/qa`	🧪 QA lead running real browser tests, filing bugs, writing regression tests
`/polish-beta`	✨ HITL polish pass for copy, microcopy, and last-mile release quality
`/ship`	🚀 Release engineer: tests, coverage audit, PR open, deploy handoff
`/land-and-deploy`	🎯 Release owner who merges, deploys, canaries, and verifies health
`/canary`	🐤 Post-deploy verifier for critical flows, console errors, and perf regressions
`/document-release`	📝 Docs steward syncing CHANGELOG, README, API docs, and migration notes
`/compound`	🧠 Knowledge curator capturing patterns for next time
`/investigate`	🔎 Debugger who traces data flow and logs hypotheses
`/technical-writing`	📚 Tech writer: README, API docs, migration guides
`/devops-pipeline`	⚙️ DevOps engineer: CI/CD, IaC, zero-downtime deploys
`/database-optimization`	🗄️ DBA: query plans, indexing, N+1 detection

Agents: Architect (designs, never implements) · Tester (edge cases, screenshot evidence) · Debugger (read-only investigator) · Sentinel (security review, 8/10+ confidence gate) · Reality Checker (evidence-based certification, default NEEDS WORK)

Design — 3 skills, 1 agent

Skill	Your specialist
`/design-system`	🎨 Design lead building tokens, components, patterns from scratch
`/design-review`	🖼️ Critic with AI slop detection, 0-10 scoring, and brand alignment
`/ux-research`	🔬 Researcher: test scripts, personas, journey maps

Agent: Design Critic (detects AI slop, scores craft, severity classification)

Security — 2 skills, 1 agent

Skill	Your specialist
`/cso`	🛡️ Security officer: OWASP + STRIDE + supply chain + zero-trust
`/benchmark`	⚡ Performance engineer: Core Web Vitals, load testing, capacity planning

Agent: Sentinel (read-only security review, 8/10+ confidence gate)

Quality & Learning — 5 skills

Skill	Your specialist
`/slop-scan`	🧹 AI slop detector: catches empty catches, redundant awaits, dead code, copy-paste smell
`/learn`	📖 Knowledge engineer: captures cross-session operational learnings
`/health`	🏥 Code health monitor: weighted 0-10 score for coverage, complexity, duplication, docs
`/scrape`	🔧 Data extractor: structured data from web pages via curl + jq
`/retro`	📊 Sprint retrospective: what worked, what didn't, data-driven action items

Workflow commands

Chain skills into end-to-end pipelines:

Command	What happens
`/autoplan`	🔍 Multi-lens plan review before `/build`
`/sprint`	💡 brainstorm → plan → build → review → qa → polish-beta → ship → land-and-deploy → document-release → compound
`/hotfix`	🔥 investigate → fix → review → ship
`/tdd`	🧪 red → green → refactor (enforced)
`/design-sprint`	🎨 research → design system → review → iterate
`/security-scan`	🛡️ OWASP + STRIDE + dependency audit

How it works

Each skill is a structured process — numbered steps, decision criteria, quality gates, explicit deliverables, and DOT flowcharts. Your agent follows the playbook instead of improvising.

skills/
├── brainstorm/SKILL.md     ← Role + Process + Rules + Deliverables + DOT Flowchart
├── plan/SKILL.md
├── build/SKILL.md
├── ...
agents/
├── architect.md            ← Persona + Expertise + Decision Framework
├── tester.md
├── reality-checker.md      ← Evidence-based certification, default NEEDS WORK
├── ...
commands/
├── sprint.md               ← Skill chain with human approval gates
├── hotfix.md
├── ...
hooks/
├── session-start           ← Universal bootstrap with feature discovery
├── hooks.json              ← SessionStart + PreToolUse + PostToolUse

Skills are written in SKILL.md with name, description, version, and triggers frontmatter — the universal format across all supported CLIs. Write once, run anywhere.

Multi-CLI support

Feature	Claude Code	Codex CLI	OpenCode
Skills	Native (marketplace)	Native (plugin)	Native (plugin)
Agents	Native	With `multi_agent`	Native (plugin)
Hooks	Full (SessionStart + Pre/Post)	SessionStart + Pre/Post	Via JS plugin
Commands	Native	Native (plugin)	Native (plugin)

Tool name differences are bridged automatically via reference docs in skills/using-solo-squad/references/.

Cost

Subagent workflows use more tokens than simple prompting.

Workflow	Typical cost
Single skill	< $1
`/sprint` (full cycle)	$5-15
`/security-scan`	$2-4

On Pro/Max/Team plans with included usage, this rarely matters.

Built on

superpowers — TDD discipline, subagent-driven development, multi-CLI architecture, DOT flowcharts, rationalization tables
gstack — Host config system, browser automation, safety guardrails, persistent memory
compound-engineering — Knowledge compounding philosophy, converter architecture
agency-agents — Evidence-based certification, SRE frameworks, code review formats, orchestration patterns

Contributing

Add a skill in your domain. Each one needs a SKILL.md with frontmatter, numbered process, explicit rules, defined deliverables, and a DOT flowchart. See CONTRIBUTING.md.

License

MIT

Name		Name	Last commit message	Last commit date
Latest commit History 38 Commits
.agents/plugins		.agents/plugins
.claude-plugin		.claude-plugin
.codex-plugin		.codex-plugin
.codex		.codex
.github/workflows		.github/workflows
.opencode/plugins		.opencode/plugins
agents		agents
clients		clients
commands		commands
hooks		hooks
scripts		scripts
skills		skills
tests		tests
.gitignore		.gitignore
AGENTS.md		AGENTS.md
CHANGELOG.md		CHANGELOG.md
CLAUDE.md		CLAUDE.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
package.json		package.json

Folders and files

Latest commit

History

Repository files navigation

Solo Squad

Install

Claude Code (Recommended)

Marketplace Install (Recommended)

Manual Install

Codex CLI

Marketplace Install (Recommended)

Manual Install (Legacy)

OpenCode

Plugin Registry Install (Recommended)

Manual Install

Updating

Uninstalling

What makes this different

See it work

Ship a feature in one command

Catch what CI misses

Get a design review that actually helps

What's inside

Development — 16 skills, 5 agents

Design — 3 skills, 1 agent

Security — 2 skills, 1 agent

Quality & Learning — 5 skills

Workflow commands

How it works

Multi-CLI support

Cost

Built on

Contributing

License

About

Topics

Resources

License

Contributing

Uh oh!

Stars

Watchers

Forks

Releases 6

Uh oh!

Contributors

Uh oh!

Languages