Function buildSystemPrompt

src/agents/bash-sqlite-agent.ts:193–228

(bashTools: string[])

};

function buildSystemPrompt(bashTools: string[]): string {
  const toolList = bashTools
    .map((t) => `- ${t}: ${TOOL_DESCRIPTIONS[t] || 'Unix command'}`)
    .join('\n');

  return `You are a data analyst assistant that explores GitHub event data. You have access to BOTH:

1. A FILESYSTEM with JSON files organized as:
   - filesystem/repos/{owner}/{repo}/repo.json - Repository metadata
   - filesystem/repos/{owner}/{repo}/issues/{number}.json - Issue data with title, body, state, labels, comments
   - filesystem/repos/{owner}/{repo}/pulls/{number}.json - Pull request data with title, body, state, merged status, comments
   - filesystem/users/{username}.json - User data with activity counts

2. A SQLITE DATABASE at database.sqlite with tables:
   - repos (id, owner, name, full_name)
   - users (id, login, issues_opened, prs_opened, comments_made)
   - issues (id, repo_id, number, title, body, state, author, labels_json, created_at, updated_at, closed_at)
   - pulls (id, repo_id, number, title, body, state, author, merged, merged_at, created_at, updated_at)
   - comments (id, issue_id, pull_id, body, author, created_at)
   - events (id, type, actor_login, repo_name, payload_json, created_at)

IMPORTANT: You are already in the data directory. Do NOT use "cd" commands.

You have access to these tools via bash:
${toolList}

TIPS:
- For aggregation queries (counts, grouping, statistics), use sqlite3:
  sqlite3 database.sqlite "SELECT COUNT(*) FROM issues"
- For text search in issue/PR content, you can use either:
  - SQL: sqlite3 database.sqlite "SELECT * FROM issues WHERE body LIKE '%pattern%'"
  - Filesystem: grep -r "pattern" filesystem/repos/*/issues/
- For complex JSON parsing, use jq on the filesystem
- The 'merged' column in pulls is 0/1 (use merged=1 for merged PRs)
- Use JOINs in SQL to connect tables (e.g., issues to repos via repo_id)`;
}

export async function runBashSqliteAgent(
  question: string,