-
Notifications
You must be signed in to change notification settings - Fork 1
Skills Reference
Skills are the core unit of capability in agent-toolkit. Each skill is a portable, self-contained definition that tells an AI coding assistant what to do in a specific situation. There are 52 skills across 9 domains.
A skill is a directory containing two files:
The human- and AI-readable prompt body. This is what the AI tool loads as its instruction set for the skill. It uses YAML frontmatter followed by Markdown content:
---
name: gh-fix-ci
description: Triage failing GitHub Actions checks and propose minimal fixes
tools: [claude-code, opencode, cursor, windsurf]
triggers:
- fix CI
- debug failing checks
- why is the build red
requires:
- gh
produces:
- plan.md
- draft PR
---
# gh-fix-ci
[Skill body here β instructions, steps, safety rules, examples]Skills use SKILL.md frontmatter only β no skill.json required. See docs/MIGRATION.md for
historical notes on retiring legacy skill.json manifests.
Validated by schemas/skill-md-frontmatter.schema.json. Required fields are marked with an asterisk.
| Field | Type | Required | Description |
|---|---|---|---|
name * |
string | Yes | Kebab-case skill identifier. Must match the directory name exactly. Pattern: ^[a-z0-9]+(-[a-z0-9]+)*$
|
description * |
string | Yes | Short summary used for skill selection by the AI. Prefix with HOW β for tool skills or WHAT β for workflow skills |
tools |
array | No | AI tools this skill is verified to work with. Values: claude-code, opencode, cursor, windsurf, copilot-cli, pi, universal
|
triggers |
array | No | Natural-language phrases that should invoke this skill |
requires |
array | No | CLI tools, commands, or env vars the skill needs |
produces |
array | No | Artifact types the skill outputs (e.g. plan.md, PR, comment) |
compatibility |
string | No | Human-readable note about caveats or prerequisites |
metadata.author |
string | No | GitHub username of the skill author |
metadata.version |
string | No | Semantic version (e.g. "1.0") |
Description prefix convention:
-
HOW βfor tool skills that know how to operate a specific CLI or API -
WHAT βfor workflow skills that know what to do at each project phase
8 skills that underpin how the AI agent operates in any session. These are loaded by default in every profile.
| Skill | Description | CC | Cursor | OC | CP | WS | Pi |
|---|---|---|---|---|---|---|---|
assistant |
General-purpose AI assistant with repo inspection and convention verification. Default entry point when no other skill matches. | Yes | Yes | Yes | Yes | Yes | Yes |
dev-companion |
Dev companion orchestration and routing. Use for general client or project delivery work when the right workflow skill is unclear. | Yes | Yes | Yes | No | Yes | Yes |
onboarding |
Guided project onboarding for new team members. Initializes workspace context, discovers conventions, and walks through the repository. | Yes | Yes | Yes | Yes | Yes | Yes |
output-handshake |
Artifact destination gate. Confirms where to save before writing any final output. Triggers on: "where to save", "final artifact", "output destination". | Yes | Yes | Yes | No | Yes | Yes |
pr-fallback |
Default PR/MR body generator when no project template exists. Structures description, motivation, testing, and checklist sections. | Yes | Yes | Yes | Yes | Yes | Yes |
workspace-knowledge-sync |
Sync workspace knowledge base, save learnings and todos, inject session context. Requires a knowledge/ directory at workspace root. |
Yes | No | Yes | No | No | No |
CC = Claude Code, OC = OpenCode, CP = Copilot, WS = Windsurf
21 skills covering the full spectrum from requirements through handoff. These skills handle PRDs, ADRs, work items, meetings, incidents, and assessments.
| Skill | Description | CC | Cursor | OC | CP | WS | Pi |
|---|---|---|---|---|---|---|---|
adr |
Architecture Decision Record lifecycle and authoring. Triggers on: ADR, architecture decision record, technical decision record. | Yes | Yes | Yes | Yes | Yes | Yes |
agreement |
Agreement document with parties, terms, conditions, and validity period. | Yes | Yes | Yes | Yes | Yes | Yes |
bug |
Bug report template with reproduction steps, expected vs. actual behavior, and severity classification. | Yes | Yes | Yes | Yes | Yes | Yes |
decision-log |
Lightweight decision log entries with rationale, alternatives considered, and reversibility. | Yes | Yes | Yes | Yes | Yes | Yes |
development-workflow |
Default task lifecycle, Definition of Ready, Definition of Done, and validation model. | Yes | Yes | Yes | Yes | Yes | Yes |
epic |
Epic template with objectives, success criteria, and scope boundaries. | Yes | Yes | Yes | Yes | Yes | Yes |
incident |
Incident report and root cause analysis template. Triggers on: incident, RCA, production outage, service degradation. | Yes | Yes | Yes | Yes | Yes | Yes |
management-unit-assessment |
Management scorecard for governance, delivery process, and AI-native management readiness. | Yes | Yes | Yes | No | Yes | No |
meeting-minutes |
Meeting minutes with agenda, key decisions, and action items. | Yes | Yes | Yes | Yes | Yes | Yes |
planning |
Planning, estimation, sprint capacity, and task breakdown. | Yes | Yes | Yes | Yes | Yes | Yes |
prd |
Product Requirements Document authoring with personas, success metrics, and constraints. | Yes | Yes | Yes | Yes | Yes | Yes |
project-assessment |
Interactive project assessment intake and scoring. Routes to technical and management unit assessments. | Yes | Yes | Yes | No | Yes | No |
project-assessment-evidence |
Evidence map for assessment sources, confidence levels, and missing evidence. | Yes | Yes | Yes | No | Yes | No |
spike |
Spike and research findings with tradeoffs, options, and recommendation. | Yes | Yes | Yes | Yes | Yes | Yes |
task |
Technical task template with acceptance criteria and implementation notes. | Yes | Yes | Yes | Yes | Yes | Yes |
technical-unit-assessment |
Technical scorecard for frontend, backend, and infrastructure readiness. | Yes | Yes | Yes | No | Yes | No |
trd |
Technical Requirements Document authoring with architecture context and constraints. | Yes | Yes | Yes | Yes | Yes | Yes |
user-story |
User story template with persona, goal, and acceptance criteria. | Yes | Yes | Yes | Yes | Yes | Yes |
work-item |
Routes epics, stories, tasks, bugs, and incidents to their atomic templates. | Yes | Yes | Yes | Yes | Yes | Yes |
workflow-client-bootstrap |
Generates client-specific workflow skill pairs for new client onboarding. | Yes | Yes | Yes | No | Yes | No |
workflow-generic-project |
Delivery phases (plan β implement β review β PR) for any client project. | Yes | Yes | Yes | Yes | Yes | Yes |
6 skills that bridge design tools and code.
| Skill | Description | CC | Cursor | OC | CP | WS | Pi |
|---|---|---|---|---|---|---|---|
figma |
Figma MCP entry point. Validates Figma MCP connection and sets up required context before any design-to-code task. Requires Figma MCP configured. | Yes | No | Yes | No | Yes | No |
figma-code-connect-components |
Maps Figma nodes to existing code components using Figma Code Connect annotations. | Yes | No | Yes | No | Yes | No |
figma-create-design-system-rules |
Generates design system rule files (AGENTS.md style) from an existing Figma file. Encodes unwritten conventions. | Yes | No | Yes | No | Yes | No |
figma-create-new-file |
Creates a new Figma file with sane defaults, naming conventions, and initial frame structure. | Yes | No | Yes | No | Yes | No |
figma-implement-design |
Translates a Figma node to production code with pixel-level fidelity. Uses Figma MCP to extract design context. | Yes | Yes | Yes | No | Yes | No |
ui-ux-pro-max |
UI/UX design, palette enforcement, component architecture, and accessibility review. Does not require Figma. | Yes | Yes | Yes | Yes | Yes | Yes |
Note: All figma-* skills require the Figma MCP provider to be configured. See MCP Setup for details.
5 skills for code delivery on hosted platforms.
| Skill | Description | CC | Cursor | OC | CP | WS | Pi |
|---|---|---|---|---|---|---|---|
github-cli-workflow |
Push branch and create draft PR with gh. Handles branch naming, PR body, and reviewer assignment. Requires gh CLI. |
Yes | Yes | Yes | No | Yes | Yes |
gitlab-cli-workflow |
Push branch and create draft MR with glab. GitLab equivalent of github-cli-workflow. Requires glab CLI. |
Yes | Yes | Yes | No | Yes | Yes |
gh-address-comments |
Inspects open PR review comments and applies fixes one by one, confirming with the user before each change. | Yes | Yes | Yes | No | Yes | Yes |
gh-fix-ci |
Triage failing GitHub Actions checks. For simple failures, proposes a minimal fix. For complex failures, posts a diagnosis comment. | Yes | Yes | Yes | No | Yes | Yes |
gh-contribution-planner |
Plans OSS contributions: finds suitable issues, evaluates complexity, and drafts a contribution strategy. | Yes | Yes | Yes | No | Yes | Yes |
Example β fix CI:
"fix CI"
"why is the build red"
"debug failing checks on this PR"
4 skills for team collaboration tools.
| Skill | Description | CC | Cursor | OC | CP | WS | Pi |
|---|---|---|---|---|---|---|---|
clickup-cli |
View ClickUp task details, add comments, and update task status via the ClickUp CLI. Triggers on task IDs with CU- prefix. |
Yes | Yes | Yes | No | Yes | No |
linear |
Manage Linear issues, projects, and cycles via Linear MCP. Create issues, add comments, query sprints and assignments. Requires Linear MCP. | Yes | No | Yes | No | Yes | No |
slack-assistant |
Read Slack channels and messages, send messages, add reactions, and browse canvases. Requires Slack MCP. | Yes | No | Yes | No | Yes | No |
slack-cli |
Slack app development: scaffold Bolt apps, manage manifests, and deploy via the Slack CLI. Different from slack-assistant. |
Yes | Yes | Yes | No | Yes | No |
When to use slack-assistant vs slack-cli:
-
slack-assistantβ reading channels, sending notifications, checking messages in a Slack workspace -
slack-cliβ building and deploying Slack apps and bots
2 skills for data engineering workflows.
| Skill | Description | CC | Cursor | OC | CP | WS | Pi |
|---|---|---|---|---|---|---|---|
dbt-validation |
Run dbt parse, dbt compile, dbt test, and dbt run according to the project's docs conventions. Triggers when dbt_project.yml is present. |
Yes | Yes | Yes | No | Yes | No |
snowflake-validation |
Read-only validation patterns for Snowflake queries and tables via CLI or SQL. | Yes | Yes | Yes | No | Yes | No |
2 skills for browser automation and notebook workflows.
| Skill | Description | CC | Cursor | OC | CP | WS | Pi |
|---|---|---|---|---|---|---|---|
jupyter-notebook |
Scaffold reproducible Jupyter notebooks from templates. Convert Python scripts to notebooks. | Yes | Yes | Yes | No | Yes | No |
playwright-cli |
Drive a real browser from the terminal via Playwright MCP. Extract data from rendered pages, reproduce UI bugs, take screenshots. | Yes | Yes | Yes | No | Yes | No |
3 operational skills.
| Skill | Description | CC | Cursor | OC | CP | WS | Pi |
|---|---|---|---|---|---|---|---|
docs-generator |
Generate documentation from code and inline comments. Update README or API docs after code changes. | Yes | Yes | Yes | Yes | Yes | Yes |
llm-cost-advisor |
Analyze and advise on LLM token costs. Estimate cost for a workflow. Compare token cost across providers. Budget loop runs. | Yes | Yes | Yes | No | Yes | No |
triage |
General-purpose issue and task triage for workstation health, environment diagnosis, and missing tool resolution. | Yes | Yes | Yes | Yes | Yes | Yes |
1 skill that manages the loop engineering system.
| Skill | Description | CC | Cursor | OC | CP | WS | Pi |
|---|---|---|---|---|---|---|---|
loop-runner |
Schedule and manage recurring agentic loops. Handles STATE.md checkpointing, cadence management, and budget enforcement. | Yes | No | Yes | No | No | No |
{
"skillPaths": [
"/path/to/agent-toolkit/skills/forge/gh-fix-ci",
"/path/to/agent-toolkit/skills/delivery/workflow-generic-project"
]
}Or install the full Claude Code profile:
cp -r ~/.agent-toolkit/profiles/claude-code/. ~/.claude/# All domains
mkdir -p .cursor/rules
cp ~/.agent-toolkit/profiles/cursor/rules/*.mdc .cursor/rules/
# One domain only
cp ~/.agent-toolkit/profiles/cursor/rules/forge.mdc .cursor/rules/Available rule files:
profiles/cursor/rules/
βββ core.mdc # Core patterns and conventions
βββ delivery.mdc # Work item lifecycle, PRD, TRD, ADR
βββ design.mdc # UI/UX and Figma integration
βββ forge.mdc # GitHub/GitLab CLI workflows
βββ integrations.mdc # Slack, Linear, ClickUp
βββ ops.mdc # Triage, incident response
cp -r ~/.agent-toolkit/profiles/opencode/. ~/.config/opencode/mkdir -p .github
cp ~/.agent-toolkit/profiles/copilot/copilot-instructions.md .github/copilot-instructions.mdEdit the file to remove domains that do not apply to your project.
cp -r ~/.agent-toolkit/profiles/windsurf/. ~/.codeium/windsurf/mkdir -p ~/.pi/agent/skills
cp -r ~/.agent-toolkit/profiles/pi/skills/. ~/.pi/agent/skills/Run before deploying or submitting a PR:
# Validate all skills (schema, frontmatter, no secrets)
bash scripts/validate-skills.sh
# Validate a specific skill using the Python validator
python3 scripts/validate-skills.pyThe validators check:
- Every skill directory has
SKILL.md -
SKILL.mdfrontmatter has requirednameanddescriptionfields - Frontmatter validates against
schemas/skill-md-frontmatter.schema.json - No deprecated
skill.jsonfiles are present -
namein frontmatter matches the directory name exactly - No secrets or credential patterns are present in any file
Getting Started
Reference
- π οΈ Skills
- π€ Agents
- π Loop Engineering
- π MCP Setup
- π₯οΈ Profiles
- π Plugin Marketplace
Compiler