Skills

Install, manage, and discover agent skills. Skills are packaged instructions that extend what your agent can do.

What Are Skills?

A skill is a directory inside $SIGNET_WORKSPACE/skills/ containing at minimum a SKILL.md file. That file describes what the skill does and how the agent should use it. Skills are markdown — they contain instructions, not executable code.

$SIGNET_WORKSPACE/skills/
├── remember/
│   └── SKILL.md          # /remember command
├── recall/
│   └── SKILL.md          # /recall command
├── browser-use/
│   └── SKILL.md          # browser automation
└── github/
    └── SKILL.md          # GitHub CLI operations

When a harness loads your agent, the installed skills are included in the system prompt (via the <available_skills> section in AGENTS.md). The agent reads the relevant SKILL.md and follows its instructions.

CLI Commands

# List installed skills
signet skill list

# Search the skills.sh registry
signet skill search browser

# Install from skills.sh
signet skill install browser-use

# Show details about a skill
signet skill info browser-use

# Remove a skill
signet skill remove weather

# Update all installed skills
signet skill update

# Update a specific skill
signet skill update github

# Create a new skill (scaffolding)
signet skill create my-tool

Built-in Skills

Signet ships with three built-in skills that integrate directly with the Memory system (see Memory Skills for full details):

`/remember`

Save information to persistent memory. Content is embedded automatically for semantic search.

# CLI
signet remember "nicholai prefers bun over npm"
signet remember "critical memory" --critical
signet remember "tagged memory" -t tag1,tag2

# In harness
/remember nicholai prefers bun over npm
/remember critical: never push directly to main
/remember [tag1,tag2]: tagged content here

Prefixes:

critical: or --critical — pins the memory (never decays, always included in session context)
[tag1,tag2]: or -t tag1,tag2 — adds searchable tags

See MEMORY.md for full details.

`/recall`

Search memories with hybrid semantic + keyword search.

# CLI
signet recall "coding preferences"
signet recall "signet architecture" --type decision -l 5

# In harness
/recall coding preferences
/recall signet architecture
/recall what did we decide about auth

See MEMORY.md for full details.

`/memory-debug`

Diagnose memory issues end-to-end (daemon health, remember/write path, recall/read path, embedding health, and API checks).

# In harness
/memory-debug
/memory-debug recall is empty

# Core checks it should run
signet status
signet remember "memory-debug smoke test" -t debug,smoke
signet recall "memory-debug smoke test" --json

Use this when memory results are missing, low quality, or inconsistent across sessions.

SKILL.md Format

Skills are defined with YAML frontmatter followed by markdown content.

```markdown

name: example-skill description: Brief description for skill listings version: 1.0.0 author: username homepage: https://github.com/user/skill-repo dependencies: bins: [some-cli] # Required CLI tools skills: [other-skill] # Required other skills user_invocable: true # Can be triggered with /skill-name arg_hint: “” # Help text for args

builtin: false # true for Signet built-ins

Skill Name

What this skill does and when to use it.

Usage

/skill-name some argument

Examples

…

Implementation

Instructions for the agent on how to carry out this skill…

The `user_invocable: true` flag means users can trigger the skill with `/skill-name` in their harness. The `builtin: true` flag marks skills that integrate directly with the Signet daemon API (like remember and recall).

---

## skills.sh Integration

[skills.sh](https://skills.sh) is a public registry of agent skills. Signet searches and installs from it via the `npx skills` CLI.

### How install works

1. `signet skill install <name>` calls `npx skills add <name> --global --yes`
2. The skill is installed to `$SIGNET_WORKSPACE/skills/<name>/`
3. The SKILL.md file is parsed for metadata
4. The harness config is updated to include the new skill

### How search works

1. `signet skill search <query>` calls `npx skills search <query>`
2. Results show name, description, and install count
3. Already-installed skills are marked

---

## Dashboard

The [Dashboard](/dashboard/)'s **Skills** panel provides a UI for everything the CLI does:

- **Installed tab** — lists skills with name, description, version
- **Browse tab** — search skills.sh, install with one click
- **Detail view** — shows the full SKILL.md content

---

## Creating Custom Skills

```bash
signet skill create my-tool

This creates $SIGNET_WORKSPACE/skills/my-tool/SKILL.md with a starter template. Edit it to describe what the skill does.

Example: a custom deployment skill

```markdown

name: deploy-staging description: Deploy the current project to staging version: 0.1.0 author: nicholai user_invocable: true

arg_hint: “[service-name]”

deploy-staging

Deploys the current project to the staging environment.

Usage

/deploy-staging
/deploy-staging api-service

Implementation

Check that you’re on a clean git branch (no uncommitted changes)
Run bun run build to build the project
Run the project’s deploy script: ./scripts/deploy.sh staging
Report the deployment URL when complete

If there are uncommitted changes, stop and ask the user if they want to commit first.

---

## Harness Integration

### Claude Code and OpenCode

An `<available_skills>` section is generated in the AGENTS.md copy for each harness, listing installed skills by name and description. The agent uses this to know which skills are available.

### OpenClaw

Skills are referenced in the `skills.entries` section of the OpenClaw workspace config, pointing to the `$SIGNET_WORKSPACE/skills/` directory.

---

## Security

Skills are markdown files — they're instructions, not code. A skill only does what the agent chooses to do after reading it. However:

- A malicious skill could instruct the agent to do harmful things.
- Only install skills from sources you trust.
- Skill signing and verification is planned for a future release.

---

## Publishing a Skill

To share a skill on skills.sh:

```bash
cd $SIGNET_WORKSPACE/skills/my-skill
signet skill publish

Requirements:

Valid SKILL.md with required frontmatter (name, description, version, author)
Unique skill name on skills.sh
skills.sh account (authenticate via signet auth)

Versioning

Skills use semver in the version frontmatter field. signet skill update compares the installed version against the registry and replaces if newer.

User-created skills (no registry source) are never auto-updated.