shape-your-agent — Claude Code skill: write your agent's system prompt
Walks you through writing a system prompt for an agent on the Sanity Context MCP server: voice, boundaries, refusal behaviour.
npx skills add sanity-io/context --skill shape-your-agent -a claude-codeWhat shape-your-agent does
This is a conversation, not a checklist. The skill starts with who talks to your agent, where they do it, and what state they arrive in — exploring, stuck, annoyed. Those answers set the tone for everything after.
Then you pick positions on voice, answer length, technical depth and formatting. Each axis comes with a weak phrasing and a stronger one, so you can see the difference. Boundaries get their own pass: every rule needs a rule, a trigger scenario and the response you want. If you cannot name a real user message that fires a rule, the skill tells you to drop it. The draft lands in a short structure — role, voice, boundaries, fallback — usually a couple hundred words.
You need control over the system prompt to get anything out of this. Content schema, field meanings and query patterns belong in the Instructions field and the MCP itself; copying them into the prompt creates conflicts. No access to the system prompt? Skip the skill — the Instructions field carries the setup on its own.
What it helps with
- replace vague tone instructions with concrete voice rules
- write refusals and redirects tied to real scenarios
- decide what the agent says when a query returns nothing
- cut rules that no user message can trigger
- keep agent behaviour separate from data concerns
Example prompts for the agent
- Help me write a system prompt for our support agent — it should never name competitors
- My agent sounds too stiff, let's run the session and rework its voice
- Set boundaries: the agent reads content but never edits the dataset
How to install shape-your-agent
1Skills CLI (Vercel)
npx skills add sanity-io/context --skill shape-your-agent -a claude-code2Manually: put the files in place
~/.claude/skills/shape-your-agent/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/shape-your-agent/
1Skills CLI (Vercel)
npx skills add sanity-io/context --skill shape-your-agent -a codex2Manually: put the files in place
.agents/skills/shape-your-agent/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add sanity-io/context --skill shape-your-agent -a cursor2Manually: put the files in place
.cursor/skills/shape-your-agent/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install sanity-io/context shape-your-agent2Manually: put the files in place
.github/skills/shape-your-agent/SKILL.md1Skills CLI (Vercel)
npx skills add sanity-io/context --skill shape-your-agent -a gemini-cli2Platform CLI
gemini skills install https://github.com/sanity-io/context --path skills/shape-your-agentManually: put the files in place
.devin/skills/shape-your-agent/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add sanity-io/context --skill shape-your-agent -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about shape-your-agent
How do I install shape-your-agent in Claude Code?
Run npx skills add sanity-io/context --skill shape-your-agent -a claude-code. The skill lands at ~/.claude/skills/shape-your-agent/SKILL.md.
What does shape-your-agent do?
It runs an interactive session that produces a system prompt for an agent powered by the Sanity Context MCP server — role, voice, boundaries and fallback behaviour.
What do I need before running it?
Access to your agent's system prompt. If you cannot edit it, skip the skill: the Instructions field works without it.
