Skills¶

Skills are reusable instruction modules that get injected into the system prompt at runtime. They let you shape agent behaviour without editing the agent's core prompt.

How Skills Work¶

Each skill has a name, instructions (free-text), and tags
Skills are installed per-workflow, not per-agent — so the same agent can behave differently in different contexts
At runtime, skill instructions are appended to the agent's system prompt before the SDK session starts

graph LR
    SP[Agent System Prompt] --> Combined[Combined Prompt]
    S1[Skill: rca-format] --> Combined
    S2[Skill: polite-tone] --> Combined
    Combined --> SDK[Copilot SDK Session]

Creating a Skill¶

curl -X POST http://localhost:8000/api/skills \
  -H "Authorization: Bearer $GITHUB_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "rca-format",
    "description": "Structures output as a Root Cause Analysis report",
    "instructions": "Structure your final output as an RCA report with sections: Summary, Timeline, Root Cause, Impact, Remediation, Prevention.",
    "tags": ["incident", "reporting"]
  }'

Installing a Skill into a Workflow¶

curl -X POST http://localhost:8000/api/workflows/<WF_ID>/skills/<SKILL_ID> \
  -H "Authorization: Bearer $GITHUB_TOKEN"

To remove a skill:

curl -X DELETE http://localhost:8000/api/workflows/<WF_ID>/skills/<SKILL_ID> \
  -H "Authorization: Bearer $GITHUB_TOKEN"

Skill Fields¶

Field	Type	Description
`name`	string	Unique name for the skill
`description`	string	Human-readable description
`instructions`	string	Free-text instructions injected into the system prompt
`tags`	string[]	Free-form labels for categorisation

Example Skills¶

RCA FormatPolite ToneJSON Only

{
  "name": "rca-format",
  "instructions": "Structure your final output as an RCA report with sections: Summary, Timeline, Root Cause, Impact, Remediation, Prevention."
}

{
  "name": "polite-tone",
  "instructions": "Always respond in a polite and professional tone. Avoid jargon unless the user uses it first."
}

{
  "name": "json-output",
  "instructions": "Always respond with valid JSON. Do not include explanation text outside the JSON block."
}