a0_agent_skills

A plugin for Agent Zero that ports the addyosmani/agent-skills development workflow toolkit -- 21 curated skills, 3 specialist agent profiles, and 7 slash commands -- into the Agent Zero plugin system. It gives any Agent Zero agent access to a production-grade software engineering lifecycle: spec → plan → build → test → review → ship, with optional telemetry to track skill activations during workflow validation.

---

Quick Start

Once the plugin is installed and enabled:

# 1. Discover the right skill for your task
skills_tool:search  query="write a spec for a login system"

# 2. Start the engineering lifecycle
/spec Build a user authentication system

# 3. Plan the work
/plan

# 4. Implement incrementally
/build Implement the login endpoint

# 5. Test with TDD
/test Add edge case tests for login failure

# 6. Review before merging
/review

# 7. Ship with a GO/NO-GO decision
/ship v1.0 authentication system

Or ask naturally -- the agent will detect the right skill:

• "Let's write a spec before we start coding" → spec-driven-development
• "Break this feature into tasks" → planning-and-task-breakdown
• "Something is broken, help me debug it" → debugging-and-error-recovery

---

Prerequisites

The commands plugin must be installed and active. The 7 slash commands provided by this plugin (/spec, /plan, /build, /review, /test, /code-simplify, /ship) rely on the commands plugin's dispatch infrastructure.

Verify it is installed:

/a0/usr/plugins/commands/

If it is not present, install it via the Plugin Hub before enabling this plugin.

---

Routing System

The plugin uses a system_prompt extension to inject mandatory routing rules into every Agent Zero session. This extension lives at:

/a0/usr/plugins/a0_agent_skills/extensions/python/system_prompt/_15_agent_skills_routing.py

It reads routing rules from:

/a0/usr/plugins/a0_agent_skills/prompts/agent.skills.routing.md

Key advantage: This approach works regardless of project context. Whether a project is active or not, the routing rules are always injected into the agent's system prompt during prompt assembly. This is the same pattern used by the built-in _memory and _browser plugins.

---

The 6-Phase Lifecycle

Every feature or change passes through six phases in order. Each phase has a required skill:

Phase	Required Skill(s)	Purpose
DEFINE	spec-driven-development	Define what to build before building it
PLAN	planning-and-task-breakdown	Break work into ordered, testable increments
BUILD	incremental-implementation + test-driven-development	Implement in slices with tests first
VERIFY	debugging-and-error-recovery	Prove it works -- fix issues before review
REVIEW	code-review-and-quality	Structured quality gate before shipping
SHIP	shipping-and-launch	Deploy with confidence and rollback plan

Skills by Phase (Full Catalog)

DEFINE: spec-driven-development

PLAN: planning-and-task-breakdown, context-engineering

BUILD: incremental-implementation, test-driven-development, source-driven-development, frontend-ui-engineering, api-and-interface-design

VERIFY: browser-testing-with-devtools, debugging-and-error-recovery

REVIEW: code-review-and-quality, code-simplification, security-and-hardening, performance-optimization

SHIP: shipping-and-launch, ci-cd-and-automation, git-workflow-and-versioning, documentation-and-adrs, deprecation-and-migration

---

The 3 Agent Profiles

Specialist subordinate agents activated via call_subordinate. These personas are invoked at the correct lifecycle phase:

code-reviewer

Senior staff engineer conducting five-axis code review: correctness, readability, architecture, security, performance. Produces structured APPROVE / REQUEST CHANGES reports with Critical / Important / Suggestion findings and file:line references.

Invoked during: REVIEW phase (via /review) and SHIP phase (via /ship).

call_subordinate(profile="code-reviewer")

security-auditor

Security engineer focused on vulnerability detection and threat modeling. Covers OWASP Top 10, input handling, authentication/authorization, data protection, infrastructure, and third-party integrations. Produces severity-classified findings (Critical / High / Medium / Low / Info) with actionable mitigations.

Invoked during: SHIP phase (via /ship).

call_subordinate(profile="security-auditor")

test-engineer

QA engineer for test strategy, test writing, and coverage analysis. Implements the Prove-It Pattern for bug reproduction. Produces Test Coverage Analysis reports with Recommended Tests and priority classification.

Invoked during: VERIFY phase and SHIP phase (via /ship).

call_subordinate(profile="test-engineer")

Orchestration rule: The main agent (or a slash command) is the orchestrator. Personas do not invoke each other. The only multi-persona pattern is parallel fan-out with a merge step, used by /ship.

Per-persona model override: Agent Zero does not support a model field directly in agent.yaml. To assign a specific LLM model to a profile (e.g. Haiku for test-engineer, Opus for security-auditor), create a _model_config/config.json alongside the profile:

/a0/usr/plugins/a0_agent_skills/agents/security-auditor/plugins/_model_config/config.json

This file must contain a complete effective configuration for chat_model, utility_model, and embedding_model (selected as a whole, not merged). See the _model_config plugin documentation for the schema.

---

The 7 Slash Commands

Command	When to use
/spec	Start a new feature or project -- loads spec-driven-development and writes SPEC.md. Surfaces assumptions before writing any code.
/plan	Break confirmed spec into tasks -- loads planning-and-task-breakdown and writes tasks/plan.md + tasks/todo.md.
/build	Implement the next task -- loads incremental-implementation + test-driven-development. RED → GREEN → commit cycle.
/test	TDD workflow or bug reproduction -- loads test-driven-development. Write failing test first; use the Prove-It Pattern for bugs.
/review	Single-perspective code review -- delegates to the code-reviewer profile via call_subordinate.
/code-simplify	Reduce complexity without changing behavior -- loads code-simplification. Applies guard clauses, splits functions, removes dead code.
/ship	Pre-launch sequential review -- runs code-reviewer → security-auditor → test-engineer one after another, then produces a GO / NO-GO decision with rollback plan.

Example usage

/spec Build a REST API for task management
/plan
/build Implement POST /tasks endpoint
/test Add tests for the task creation edge cases
/review
/ship v1.0 release -- task API

---

Installation

Option A -- Copy directly:

cp -r /path/to/a0_agent_skills /a0/usr/plugins/a0_agent_skills

Option B -- Plugin Hub: Search for a0_agent_skills in the Agent Zero Plugin Hub and click Install.

After installation, restart Agent Zero (or use the UI restart button) so the plugin is picked up.

---

Enabling Telemetry (Phase 1 Validation)

Telemetry is off by default. Enable it to log every skills_tool activation to a JSONL file for workflow analysis.

Step 1: Copy the plugin config to your project:

mkdir -p /a0/usr/projects/<your-project>/.a0proj/plugins/a0_agent_skills/
cp /a0/usr/plugins/a0_agent_skills/default_config.yaml \
   /a0/usr/projects/<your-project>/.a0proj/plugins/a0_agent_skills/config.yaml

Step 2: Edit the config and set telemetry_enabled: true:

telemetry_enabled: true
telemetry_log_path: .a0proj/skill_activations.jsonl

Step 3: Run your workflow. Each skills_tool:load, skills_tool:search, or skills_tool:list call appends one JSON line to .a0proj/skill_activations.jsonl:

{"ts": 1234567890.0, "tool": "skills_tool:load", "skill_name": "incremental-implementation", "query": null, "result_preview": "Skill loaded..."}

Telemetry is fully guarded -- a logging failure never interrupts agent operation.

---

Skill Override: Customizing at Project Scope

You can override any skill for a specific project without modifying the plugin. Copy the skill's SKILL.md to your project's .a0proj directory:

mkdir -p /a0/usr/projects/<your-project>/.a0proj/skills/<skill-name>/
cp /a0/usr/plugins/a0_agent_skills/skills/<skill-name>/SKILL.md \
   /a0/usr/projects/<your-project>/.a0proj/skills/<skill-name>/SKILL.md

Then edit the copy. Agent Zero's skill resolution searches project scope first, so your local version takes precedence over the plugin version for that project only.

---

The 20-Skill Cap and using-agent-skills

Agent Zero's skill system works best when the agent loads skills intentionally rather than pinning everything into the system prompt permanently. The using-agent-skills skill explains when and how to invoke other skills -- it is a meta-skill designed for on-demand use.

Do not pin using-agent-skills as a persistent/always-loaded skill. It should be loaded on demand only (e.g. when onboarding a new agent to the workflow).

If you are pinning skills for a project, prioritize the 7 lifecycle skills. Stay within the 20-skill context budget to maintain performance.

---

Attribution

Skills, agent personas, and slash commands ported from addyosmani/agent-skills by Addy Osmani. Adapted for the Agent Zero plugin system.