> ## Documentation Index
> Fetch the complete documentation index at: https://doc.codika.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Creating Agent Skills

> Step-by-step guide to creating skills that make your use case workflows accessible to AI agents

## Overview

This guide walks you through adding agent skills to an existing use case. By the end, your HTTP endpoints and scheduled workflows will be discoverable and usable by any Claude-based agent.

<Info>
  **New use cases get skills automatically.** When you run `codika init`, the scaffolded use case includes example skills for both the HTTP workflow and the scheduled workflow. This guide is for adding skills to existing use cases or customizing the defaults.
</Info>

<Tip>
  **Prefer automation?** The [Builder System](/builder/overview) agents create use cases with correctly configured agent skills out of the box. Use this guide when you need to add or customize skills manually.
</Tip>

## Step 1: Identify triggerable workflows

Open your `config.ts` and list all workflows with `http` or `schedule` trigger types. Skip sub-workflows, data ingestion, and service event workflows.

```bash theme={null}
# Quick check — list workflow files that are NOT sub-workflows
grep -l '"type": "n8n-nodes-base.webhook"' workflows/*.json
grep -l '"type": "n8n-nodes-base.scheduleTrigger"' workflows/*.json
```

For each triggerable workflow, note:

* `workflowTemplateId` from config.ts
* Trigger type (http or schedule)
* Input schema fields (for HTTP)
* Output schema fields
* Integration UIDs used
* Credit cost

## Step 2: Create the skills directory

```bash theme={null}
mkdir -p skills
```

For each triggerable workflow, create a subdirectory:

```bash theme={null}
mkdir skills/main-workflow
mkdir skills/scheduled-report
```

## Step 3: Write SKILL.md files

### For HTTP workflows

Create `skills/{name}/SKILL.md`:

```markdown theme={null}
---
name: {usecase-slug}-{action}
description: {Third-person description. What it does and what it returns.}
workflowTemplateId: {workflow-template-id}
---

# {Workflow Name}

{One-line description with integrations used.}

## How to trigger

\`\`\`bash
codika trigger {workflow-template-id} --payload-file - <<'EOF'
{
  "field1": "example value",
  "field2": 42
}
EOF
\`\`\`

## Input

| Parameter | Type | Required | Description |
|-----------|------|----------|-------------|
| field1 | string | yes | What this field does |
| field2 | number | no | What this field does |

## Output

\`\`\`json
{
  "result": "example output",
  "timestamp": "2025-01-01T00:00:00.000Z"
}
\`\`\`

## Notes

- Cost: {N} credit(s)
- Uses: {Integration1}, {Integration2}
```

### For scheduled workflows

```markdown theme={null}
---
name: {usecase-slug}-{action}
description: {Third-person description. Mention the schedule AND manual trigger.}
workflowTemplateId: {workflow-template-id}
---

# {Workflow Name}

Runs automatically on a schedule ({schedule description}).

## Schedule

`{cron expression}` — {Human-readable} ({timezone})

## Manual trigger (for testing)

\`\`\`bash
codika trigger {workflow-template-id}
\`\`\`

No payload required.

## Output

| Field | Type | Description |
|-------|------|-------------|
| field1 | string | Description |

## Notes

- Cost: {N} credit(s)
- Uses: {Integration1}, {Integration2}
```

## Step 4: Validate

```bash theme={null}
codika verify use-case .
```

Check for skill-related findings:

* `SKILL-STRUCTURE` — Missing SKILL.md in a skills subdirectory
* `SKILL-NAME-FORMAT` — Name too long, wrong characters, or reserved words
* `SKILL-WORKFLOW-REF` — workflowTemplateId doesn't match any workflow
* `SKILL-DUPLICATE` — Duplicate names or workflowTemplateIds

## Step 5: Deploy

```bash theme={null}
codika deploy use-case .
```

Skills are automatically collected and sent with the deployment. No config.ts changes needed.

## Step 6: Verify agent access

```bash theme={null}
# Download the skills you just deployed
codika get skills

# Or check the JSON output
codika get skills --json
```

You should see all your skills listed with their names, descriptions, and content.

## Naming rules

The `name` field must follow Claude's naming constraints:

| Rule                                     | Example                                 |
| ---------------------------------------- | --------------------------------------- |
| Max 64 characters                        | `wat-event-weekly-digest` (24 chars)    |
| Lowercase letters, numbers, hyphens only | `propale-generate-proposal`             |
| No spaces or underscores                 | `my-skill` not `my_skill` or `my skill` |
| No reserved words                        | Cannot contain "anthropic" or "claude"  |

**Convention:** Prefix with your use case slug: `wat-`, `propale-`, `crm-`.

## Description rules

| Rule                  | Good                                                        | Bad                     |
| --------------------- | ----------------------------------------------------------- | ----------------------- |
| Third person          | "Sends a message to..."                                     | "Use this to send..."   |
| Non-empty             | "Generates a weekly digest"                                 | ""                      |
| Max 1024 characters   | Keep it concise                                             | Don't write a paragraph |
| Include what AND when | "Sends reminders daily at 8 AM. Can be manually triggered." | "Sends reminders"       |

## Progressive disclosure for complex skills

If a skill needs more than 500 lines, split into referenced files:

```
generate-proposal/
├── SKILL.md              # Overview + trigger command + basic I/O
├── INPUT-REFERENCE.md    # Detailed field-by-field input schema
└── EXAMPLES.md           # Multiple usage examples with different payloads
```

Reference them from SKILL.md:

```markdown theme={null}
For detailed input schema, see [INPUT-REFERENCE.md](INPUT-REFERENCE.md).
For more examples, see [EXAMPLES.md](EXAMPLES.md).
```

Claude reads SKILL.md first, then loads referenced files only when needed.

## Checklist

Before deploying:

* [ ] One skill per triggerable workflow (HTTP + scheduled with manual trigger)
* [ ] No skills for sub-workflows or data ingestion
* [ ] Each skill is a `{name}/SKILL.md` directory
* [ ] Frontmatter has `name`, `description`, `workflowTemplateId`
* [ ] `name` is valid (lowercase, hyphens, max 64 chars, no reserved words)
* [ ] `description` is third person and under 1024 chars
* [ ] `workflowTemplateId` matches a workflow in config.ts
* [ ] Body includes `codika trigger` command with example payload
* [ ] Body includes input/output schemas
* [ ] `codika verify use-case .` passes