Skills Catalog

Skills are workflow scripts that teach your agent how to handle specific situations. When you say “debug this” or “create a plan”, the agent matches your intent to a skill and follows its steps.

Every scaffolded agent ships with 7 essential skills. Another 29 optional skills are available to install on demand (36 total).

Essential skills (included by default)

These ship with every scaffolded agent:

Skill	What it does
agent-guide	Responds to “what can you do” — lists capabilities, tools, and how to use them
agent-persona	Activates the agent’s persona on greeting. Maintains character through the session and compaction
vault-navigator	Searches the knowledge base for patterns, prior art, and best practices
vault-capture	Persists a single known pattern, decision, or principle to the vault
systematic-debugging	First response to any bug — diagnoses root cause before proposing fixes
writing-plans	Creates structured implementation plans from clear requirements
context-resume	Rebuilds working context on session start — “where did I leave off?”

Optional skills

Install any of these to extend your agent’s capabilities:

Skill	What it does
agent-dev	Extends the agent itself — adding facades, tools, vault ops, or new skills
agent-issues	Creates GitHub issues, bugs, tasks, and milestones structured for AI agent execution
brain-debrief	Explores brain-learned patterns — strength scores, intelligence reports
brainstorming	Open-ended creative exploration when requirements are not yet clear
code-patrol	Reviews code against the project’s own vault patterns and conventions
deep-review	In-depth code review — architecture fitness, code smells, optimization opportunities
deliver-and-ship	Pre-delivery quality gates — stability, knowledge capture, code quality checks
discovery-phase	Structured exploration before committing to a plan — options, tradeoffs, recommendations
env-setup	Sets up, fixes, or restores local dev environments across languages and tools
executing-plans	Executes an approved plan step by step with review checkpoints
finishing-a-development-branch	Finalizes a branch for merge — PR preparation, cleanup, final checks
fix-and-learn	Applies a fix after root cause is found, then captures the learning in the vault
health-check	Read-only health assessment of the knowledge base — scoring, diagnostics, issues
knowledge-harvest	Extracts multiple patterns from a source — code, docs, PRs, articles
mcp-doctor	Diagnoses and repairs MCP server connectivity issues
onboard-me	Instant project orientation for newcomers — patterns, conventions, architecture overview
parallel-execute	Executes independent plan tasks concurrently using subagents
release	Bumps all monorepo package versions, commits, tags, and pushes to trigger CI/CD release
retrospective	Time-bound reflection — sprint retros, weekly summaries, actionable improvements
second-opinion	Decision support from all sources — vault, brain, cross-project experience, web research
subagent-driven-development	Decomposes tasks into independent units for parallel isolated execution
test-driven-development	Write failing tests before implementation — RED/GREEN/REFACTOR cycle
using-git-worktrees	Safe parallel branch work using git worktrees — create, work, merge, clean up
vault-curate	Knowledge maintenance — deduplicate, merge, resolve contradictions, groom
vault-smells	Deep knowledge quality analysis — contradictions, stale patterns, orphans, decay
verification-before-completion	Internal quality gate before claiming a task is done — run tests, check output
terse	Word-budget compression — ~66% output token reduction at 7.1/10 quality (benchmarked)
compress	Compresses natural language files (CLAUDE.md, memory) to reduce input tokens
yolo-mode	Autonomous execution — skip approval gates while preserving safety invariants

Installing optional skills

Install a single skill:

npx @soleri/cli skills install deep-review

Install a skill pack (group of related skills):

npx @soleri/cli pack install my-skills-pack

List installed skills:

npx @soleri/cli skills list

Choosing a skills filter

When scaffolding an agent, you can control which skills are included:

Filter	What it includes
essential	7 core skills (default)
all	All 36 skills
Custom array	Only the skills you list

Set this in your agent.yaml:

engine:
  skillsFilter: essential    # or 'all', or ['vault-navigator', 'deep-review', ...]

Skill trust levels

Every skill gets a trust level based on what files it contains. The trust classifier walks the skill directory, looks at each file’s extension, and picks the highest-risk category it finds.

Trust level	What it means	Triggered by
markdown_only	Pure documentation, no executable code	Only .md / .mdx files
assets	Contains non-markdown files but no scripts	Images, JSON, configs (no code)
scripts	Contains executable code	.sh, .ts, .js, .mjs, .cjs, .py, .rb, .bash files

The classifier escalates: if a skill directory has even one .ts file that isn’t a declaration file (.d.ts), the whole skill is classified as scripts. Declaration files are treated as references, not executable code.

Each file in the skill also gets a per-file kind:

Kind	Examples
skill	SKILL.md (the primary skill definition)
reference	.md files, .d.ts declaration files
script	.ts, .js, .sh, .py, etc.
asset	Everything else (images, JSON, configs)

To see the trust level for your installed skill packs, use the --trust flag:

npx @soleri/cli skills list --trust

This shows the trust classification, source, and engine compatibility for each pack. The output looks something like:

  my-skills@1.0.0
    skills: deep-review, code-patrol
    trust: markdown_only  source: npm  compat: compatible

Trust levels help you understand what a skill pack can do before you install it. A markdown_only skill is just workflow instructions for your AI editor. A scripts skill can run code on your machine. That distinction matters when you’re installing packs from third-party sources.

How skills work

Skills are markdown files installed to ~/.claude/skills/<name>/SKILL.md (see Your Agent for the full file-tree layout). Each skill has:

• Trigger conditions: phrases and intents that activate it
• Steps: a structured workflow the agent follows
• Tool references: which agent tools to use at each step

When the agent detects a matching intent, it loads the skill and follows the workflow. Skills compose with other agent capabilities. A skill can search the vault, create plans, capture knowledge, and use any tool available to the agent.

---

Next: Domain Packs. See also Creating Packs to build your own, Your Agent for the agent anatomy overview, Extending Your Agent for custom ops, and the CLI Reference for soleri skills and soleri pack commands.