Skills

Skills package prompts + optional scripts + optional dependencies into a bundle the agent activates on demand. Sixteen ship built-in; you can add workspace or global ones.

Built-in catalog

Skill	Category	Purpose
AWS	Cloud	EC2, S3, Lambda, CloudWatch, IAM
Datadog	Monitoring	Metrics, logs, traces, monitors
GitHub	Version control	Issues, PRs, Actions, branches, releases
Jira	Project management	Issues, sprints, updates
Kubernetes	Infrastructure	Pods, deployments, services, logs
PostgreSQL	Database	SQL, schema, backups
Sentry	Monitoring	Errors, triage, release health

Plus: Elasticsearch, Email, GitLab, Gmail, Linear, MongoDB, MySQL, Redis, Telegram.

Discovery hierarchy

Three sources checked in order. Higher priority overrides lower.

Priority	Source	Path	Scope
1 (top)	Workspace	`.codebuddy/skills/`	Project-specific
2	Global	`~/.codebuddy/skills/`	User-wide (all workspaces)
3	Built-in	Bundled with extension	Default catalog

Install

Via command palette:

CodeBuddy: Install Skill

Or ask the agent:

Install the GitHub skill so I can manage issues

Dependencies

Skills can declare CLI deps (e.g., aws, kubectl, psql). SkillService installs cross-platform:

Platform	Package managers
macOS	brew, npm, pip
Linux	apt, snap, npm, pip
Windows	choco, winget, npm, pip

Install results cached 30 s to avoid repeated subprocess calls. Architecture-specific binaries (x64, arm64) auto-selected.

Install commands go through modal approval — the full command is displayed verbatim before it runs. Invariant H4.

Environment isolation

Skills run with a restricted env. Sensitive vars blocked: LD_PRELOAD, DYLD_INSERT_LIBRARIES, PATH, HOME, SHELL, SSH_AUTH_SOCK, GPG_AGENT_INFO, HISTFILE, and others (14 total).

Using skills

Once installed, ask normally:

Create a Jira ticket for the login bug we just fixed
Show me the Redis cache hit rate from Datadog for the last hour
Deploy the latest image to the staging Kubernetes cluster

The LLM matches the request to a skill’s description, activates it (with user approval), and proceeds.

Custom skills

Create a directory under .codebuddy/skills/ with a SKILL.md:

.codebuddy/skills/
  my-skill/
    SKILL.md          # Required
    scripts/          # Optional
    install.sh        # Optional

SKILL.md — YAML frontmatter + Markdown body:

---
name: api-auditor
displayName: API Auditor
description: Audit and test API endpoints for reliability
category: testing
version: 1.0.0
dependencies:
  - name: curl
    check: curl --version
---

# API Auditor

When this skill is active, you audit API endpoints by...

Frontmatter fields:

Field	Required	Purpose
`name`	Yes	Unique identifier
`displayName`	No	Human-readable name for the UI
`description`	Yes	When to activate — matched against user requests
`icon`	No	Emoji or icon identifier
`category`	No	`cloud`, `database`, `monitoring`, …
`version`	No	Semver
`dependencies`	No	CLI tools needed (with check commands)
`config`	No	Configuration fields the skill reads
`auth`	No	Authentication type — `api-key`, `oauth`, `cli`

Full tutorial: Skills API reference.

Skills API — walkthrough of building a skill
MCP — alternative for connecting external tool servers