An AI agent skill for ParadeDB - Elasticsearch-quality full-text search in Postgres.

This skill uses a pointer-based approach. Instead of bundling static docs that can become stale, it instructs agents to fetch current ParadeDB docs from https://docs.paradedb.com/llms-full.txt at runtime and reuse the first successful fetch for the rest of the session.

npx skills add paradedb/agent-skills

npx skills add is the most stable installation path because the installer keeps up with agent-specific directory conventions.

Easiest way to sync across agents: Use dotagents to manage your skills in one place.

Use this path when npx skills add is unavailable.

Install the skill in the directory that matches your agent:

SKILL_REF="main"  # Replace with a tag or commit SHA to pin a specific version.
TARGET_DIR="$HOME/.claude/skills/paradedb-skill"  # Change for your agent.

mkdir -p "$TARGET_DIR"
curl -fsSL \
  "https://raw.githubusercontent.com/paradedb/agent-skills/${SKILL_REF}/SKILL.md" \
  -o "$TARGET_DIR/SKILL.md"
curl -fsSL \
  "https://raw.githubusercontent.com/paradedb/agent-skills/${SKILL_REF}/EXAMPLES.md" \
  -o "$TARGET_DIR/EXAMPLES.md"

For project-local installs, change TARGET_DIR to the corresponding project directory (for example, .claude/skills/paradedb-skill).

GitHub Copilot supports repository-wide instructions via .github/copilot-instructions.md.

mkdir -p .github
curl -fsSL \
  https://raw.githubusercontent.com/paradedb/agent-skills/main/SKILL.md \
  | awk 'NR == 1 && $0 == "---" { in_frontmatter = 1; next } in_frontmatter && $0 == "---" { in_frontmatter = 0; next } !in_frontmatter { if (!started && $0 == "") { next } started = 1; print }' \
  > .github/copilot-instructions.md
printf '\n\n' >> .github/copilot-instructions.md
curl -fsSL \
  https://raw.githubusercontent.com/paradedb/agent-skills/main/EXAMPLES.md \
  >> .github/copilot-instructions.md

This keeps the file in the location Copilot looks for repository instructions and strips the YAML frontmatter from SKILL.md before appending the examples.

Once installed, the skill activates when you ask your AI agent about:

• ParadeDB
• BM25 indexing
• Full-text search in Postgres
• Elasticsearch alternatives for Postgres

On the first ParadeDB question in a session, the agent should fetch live docs before answering. After a successful fetch, it should reuse that session copy for later ParadeDB questions and only refetch when the user asks for a refresh, the answer depends on newer/current changes, the earlier fetch was incomplete, or the session context is no longer available.

If docs are unavailable due to network or access errors, the skill requires the agent to report the error. When a cached session copy exists, the agent should say it can continue from that cached copy; otherwise it should ask whether to continue with local context only.

See EXAMPLES.md for categorized prompt examples.

• ParadeDB Documentation
• ParadeDB AI Agents Guide
• LLM-Optimized Docs
• ParadeDB GitHub

MIT License. See LICENSE for details.