web-search
Lightweight web search + fetch using only shell tools (curl + grep/sed/awk), for when codex exec has no built-in --search.
When & Why to Use This Skill
This lightweight Claude skill enables terminal-based web searching and content fetching using standard shell tools like curl, grep, and awk. It is specifically designed for AI agents operating in environments without built-in search capabilities, allowing them to retrieve authoritative documentation, technical references, and real-time information directly from the command line with minimal overhead.
Use Cases
- Technical Documentation Retrieval: Quickly find and cite official API specifications, manpages, or library documentation when local resources are insufficient.
- Troubleshooting & Debugging: Search for specific error messages, stack traces, or GitHub issues to identify community-driven fixes and implementation workarounds.
- Fact-Checking & Standards Verification: Verify technical standards (e.g., POSIX, NO_COLOR) or find authoritative links to support technical writing and architectural decisions.
- Code Implementation Research: Look for best-practice examples and 'cool implementations' of shell scripts or CLI tools from trusted sources like GitHub READMEs and technical wikis.
| name | web-search |
|---|---|
| description | Lightweight web search + fetch using only shell tools (curl + grep/sed/awk), for when codex exec has no built-in --search. |
| short-description | Shell-only web search helpers |
Web Search (shell-only)
Goal
Enable the agent to quickly find and cite external references from the terminal when codex exec does not support a built-in --search flag.
When to use
Use this skill when:
- the user asks you to “search the web”, “find references”, “look for cool implementations”, or “check docs online”
- the agent needs a small number of authoritative links to proceed (standards, manpages, repos, issues, docs)
Do not use it when local repo search (rg) is sufficient.
Constraints
- Uses only shell tools (
sh,curl,awk,sed,grep). - Requires network access from the environment; if blocked, fail gracefully and say so.
- Keep result sets small (default 5–10).
Workflow
- Search for candidate links:
sh .bilu/skills/web-search/scripts/web-search.sh "<query>"
- Fetch a small excerpt for context (optional):
sh .bilu/skills/web-search/scripts/web-fetch.sh "<url>"
- Prefer high-signal sources:
- official docs / specs
- upstream repos (README, docs, issues)
- no-color.org, ShellCheck wiki, shfmt docs, etc. (for shell guidance)
Scripts
scripts/web-search.sh: queries DuckDuckGo Lite and prints a list of URLs (and optionally titles).scripts/web-fetch.sh: downloads a URL and prints a trimmed, text-only excerpt (best-effort).
Examples
sh .bilu/skills/web-search/scripts/web-search.sh "bash tui fff stty read -rsn1 arrow keys"sh .bilu/skills/web-search/scripts/web-search.sh "NO_COLOR standard" --n 5sh .bilu/skills/web-search/scripts/web-fetch.sh "https://no-color.org/"