dz recommend

dz recommend takes a task description and produces three categories of recommendations: individual skills, preset bundles, and npx toolkits (full pipelines). It decomposes the task into topics, matches them against the skill/preset/toolkit knowledge bases, and falls back to learned patterns from dz pretrain / dz teach when the task is too generic.

Basic Usage

# Recommend for a specific task
dz recommend "set up CI/CD with security scanning"

# Another task
dz recommend "audit the auth module for vulnerabilities"

dz recommend "build a defi swap integration"

dz recommend takes a single quoted task description as a positional argument. It has no flags.

How Recommendations Work

dz recommend matches the task against 30 topic dimensions (the TOPIC_KEYWORDS map). The topic keys are:

Group	Topics
Build / engineering	api, testing, ci-cd, security, database, debug, frontend, git, review, monorepo
Infrastructure	kubernetes, docker, terraform, monitoring, incident
Data & integrations	data, search, email, productivity, scrape, docs, social
Domain	web3, design-thinking, product, academic, quality, health, research

Each topic carries a set of keywords (for example security → security, audit, vulnerability, owasp, injection, auth, codeql, sast). Matched topics drive the skill, preset, and toolkit recommendations.

Output Structure

A typical dz recommend output has three sections:

1. Skills

Individual skills that match the task topics:

SKILL RECOMMENDATIONS
─────────────────────
  terraform        topic: terraform
  github-actions   topic: ci-cd
  security-audit   topic: security

2. Presets

Preset bundles that cover the topics:

PRESET RECOMMENDATIONS
──────────────────────
  devops           27 skills   (covers ci-cd, terraform, kubernetes, observability)
  qe-engineer      20 skills   (covers testing, coverage, quality)

3. Full Pipelines (npx toolkits)

Suggested npx toolkits for deeper workflows (from TOOLKIT_KB):

FULL PIPELINE (npx toolkits)
────────────────────────────
  agentic-qe              npm install -g agentic-qe && aqe init --auto
  skills-feature-adr      npx @dzhechkov/skills-feature-adr init
  skills-bto              npx @dzhechkov/skills-bto init
  keysarium               npx @dzhechkov/keysarium init

Pretrain Fallback on Generic Tasks

When the task is too generic to match distinctive topics, dz recommend falls back to patterns learned from dz pretrain. This ensures useful recommendations even without explicit topic signals.

# A vague task
dz recommend "help me improve the project"
# → uses pretrain patterns + any manual dz teach patterns

If neither pretrain data nor topic signals are available, run dz pretrain first to bootstrap recommendations.

Learning Feedback

Recommendations improve over time as you teach patterns. To accelerate learning, teach patterns explicitly:

# Signal that your project follows a specific convention
dz teach "we use vitest not jest" --domain testing
dz teach "all APIs are OpenAPI 3.1 spec-first" --domain api

# Re-run recommendations
dz recommend "add API integration tests"

Examples

# CI/CD with security
dz recommend "set up CI/CD with security scanning"

# Quality engineering
dz recommend "increase test coverage and catch flaky tests"

# Research / product
dz recommend "do product discovery for a new feature"

dz pretrain — bootstrap intelligence from your codebase
dz teach — teach explicit patterns
dz scout — discover new skills from external sources
dz init --preset — install a recommended preset