SearXNG Web Search Skill

A skill for searching the web and fetching pages via a self-hosted SearXNG instance.

Three entry-point scripts cover the main use cases:

Environment Setup

All scripts read configuration from a .env file (or real environment variables).

Copy _env → .env and fill in your values:

cp _env .env

.env reference

> CLI override — every script now also accepts --searxng-url to override the env value

> for one-off runs without editing .env.

Quick Start

Install dependencies

pip install requests beautifulsoup4 lxml python-dotenv

Install PATH shims (recommended — fixes CWD issues in OS)

bash ~/.openclaude/skills/searxng-websearch/install.sh
# reload shell, then:
wsearch   "Claude Sonnet 4 release notes" --format agent
wfetch    https://example.com --max-chars 3000
wresearch "transformer attention mechanisms" --fetch-top-pages 3

Or call directly with full path (no install needed)

SKILL=~/.openclaude/skills/searxng-websearch
python3 "$SKILL/websearch.py"    "Claude Sonnet 4 release notes" --format agent
python3 "$SKILL/fetch_page.py"   https://example.com --max-chars 3000
python3 "$SKILL/deep_research.py" "transformer attention mechanisms" --fetch-top-pages 3

Script Reference

websearch.py

python websearch.py <query> [options]

Options:
  --searxng-url URL       SearXNG base URL (overrides SEARXNG_URL env var)
  --category CATEGORY     general | images | news | science | files |
                          social_media | map | music | videos | it
                          (default: general)
  --max-results N         Number of results to return (default: 5)
  --language LANG         Language code, e.g. en, de, fr
  --safe-search 0|1|2     Safe-search level
  --page N                Result page number (default: 1)
  --time-range RANGE      day | week | month | year
  --format FORMAT         markdown | text | json | agent (default: markdown)

fetch_page.py

python fetch_page.py <url> [options]

Options:
  --searxng-url URL       Unused here but accepted for consistency
  --max-chars N           Max characters to print (default: 5000)

deep_research.py

python deep_research.py <topic> [options]

Options:
  --searxng-url URL       SearXNG base URL (overrides SEARXNG_URL env var)
  --max-results N         Results per sub-query (default: 4)
  --fetch-top-pages N     Number of top pages to fetch (default: 3)
  --max-chars N           Max chars per fetched page (default: 2500)
  --output FORMAT         markdown | text (default: markdown)

Source Ranking (deep_research.py)

Pages are scored before fetching so the most authoritative content is prioritised:

⚠️ Critical: How to Invoke These Scripts (OpenClaude / Claude Code)

NEVER use cd before calling a script. Each Bash() call spawns a fresh shell;

cd skill-dir && python3 script.py silently resets the CWD and the script never runs.

Always call scripts by their full absolute path in a single command:

# ✅ CORRECT — full path, no cd
python3 ~/.openclaude/skills/searxng-websearch/websearch.py "my query" --format agent

# ✅ CORRECT — SKILL_DIR variable makes it readable
SKILL_DIR=~/.openclaude/skills/searxng-websearch
python3 "$SKILL_DIR/websearch.py" "my query" --format agent

# ❌ WRONG — cd resets on the next Bash() call
cd ~/.openclaude/skills/searxng-websearch && python3 websearch.py "my query"

If the skill path is unknown, resolve it first:

SKILL_DIR=$(find ~/.openclaude/skills -name "websearch.py" -printf '%h' -quit 2>/dev/null \
            || find ~/skills -name "websearch.py" -printf '%h' -quit 2>/dev/null)
python3 "$SKILL_DIR/websearch.py" "my query" --format agent

Tips for Claude

• Run websearch.py with --format agent when you need compact, token-efficient context to pass

back to the model.

• For broad topics, prefer deep_research.py — it fans out into sub-queries automatically.
• If SearXNG is unreachable, deep_research.py exits with a clear error message; check that

SEARXNG_URL is correct and the instance is running.

• Pipe markdown output into a file for later use:

```bash

python3 ~/.openclaude/skills/searxng-websearch/deep_research.py "RAG retrieval strategies" > research.md

```