CtrlK
BlogDocsLog inGet started
Tessl Logo

pantheon-ai/opencode-toolkit

Complete toolkit for configuring and extending OpenCode: agent creation, custom slash commands, configuration management, plugin development, and SDK usage.

75

Quality

94%

Does it follow best practices?

Impact

No eval scenarios have been run

SecuritybySnyk

Passed

No known issues

Overview
Quality
Evals
Security
Files

scenario-01.mddesign-agents/evals/

Scenario 01: Create a Specialized SQL Expert Agent

User Prompt

Create an OpenCode agent called "sql-expert" that specializes in SQL query writing and optimization. The agent should:

  1. Have a description that includes concrete trigger phrases users would say
  2. Have a system prompt written in second person
  3. Have restricted permissions — only allow reading files and running psql and explain analyze commands; block all other bash commands

Expected Behavior

  1. Write a description field that includes specific user-facing trigger phrases (e.g., "write a query", "optimize SQL", "explain schema")
  2. Write the system prompt in second person: "You are..." and "You must..." — not "The agent should..." or "This assistant..."
  3. Set bash permission to '*': 'ask' for unknown commands — not '*': 'allow'
  4. Whitelist specific safe commands (psql, explain analyze) while blocking everything else
  5. Place the agent file at .opencode/agent/sql-expert.md

Success Criteria

  • Description has concrete trigger phrases: The description field includes specific phrases like 'write a query', 'optimize SQL', 'explain schema', etc.
  • System prompt uses second person: System prompt uses 'You are...' and 'You must...' — NOT 'The agent should...' or 'This assistant...'
  • Permission uses ask for unknown bash: Permission config uses '*': 'ask' (not '*': 'allow') for bash commands
  • Specific bash commands whitelisted: Specific safe commands (psql, explain analyze) are allowed, not all commands
  • Correct file location noted: Agent file is placed in .opencode/agent/sql-expert.md

Failure Conditions

  • Writes a vague description like "A helpful SQL assistant" with no concrete trigger phrases
  • Writes the system prompt in third person ("The agent should check..." or "This assistant...")
  • Sets bash permission to '*': 'allow', granting unrestricted shell access
  • Does not whitelist specific safe commands, either allowing all or blocking all
  • Places the agent file outside .opencode/agent/

CHANGELOG.md

tile.json