x-search-faster - SKILL.md Agent Skill

name: x-search-faster description: Search X (Twitter) through Hermes Agent's Python `tools.x_search_tool` directly instead of prompting the Hermes UI/agent loop. Use when the user asks to search X/Twitter, collect recent posts, inspect X discussion, retrieve post URLs, post text, like counts, authors, handles, dates, or source links, especially when they mention `x_search`, Hermes, Grok X Search, `uvx --from hermes-agent`, or want faster/rawer X search output.

X Search Faster

Use Hermes Agent's Python x_search_tool directly:

uvx --from hermes-agent python <this-skill-dir>/scripts/run_x_search.py "SEARCH QUERY"

This avoids asking Hermes to interpret the result again. The output is still Grok's X Search analysis, not the official X API. Treat it as sourced search synthesis, and verify important posts by opening returned URLs when accuracy matters.

Required Output

For every X search result, require and preserve these fields:

投稿者: display name and @handle
いいね数: numeric like count when available, otherwise 不明
投稿本文: the visible post text, not only a summary
投稿URL: direct https://x.com/.../status/... link
投稿日: timestamp/date when available
外部リンク: URLs included in the post body, or なし

If any result lacks a post URL or post text, rerun with a stricter query or mark the item incomplete. Do not present source-free X claims as findings.

Quick Start

Run a search and print only the Markdown answer:

uvx --from hermes-agent python <this-skill-dir>/scripts/run_x_search.py \
  "今Xで話題のAIエージェント関連投稿を10件"

Limit result count:

uvx --from hermes-agent python <this-skill-dir>/scripts/run_x_search.py \
  "xAIについて最近何が話題か" --count 8

Constrain by handle or date in the augmented prompt:

uvx --from hermes-agent python <this-skill-dir>/scripts/run_x_search.py \
  "xAI product updates" --handle xai --since 2026-05-01 --until 2026-05-20

Print the raw JSON returned by Hermes:

uvx --from hermes-agent python <this-skill-dir>/scripts/run_x_search.py \
  "What are people saying about xAI on X?" --raw-json

Pass experimental x_search_tool parameters as JSON if the installed Hermes version supports them:

uvx --from hermes-agent python <this-skill-dir>/scripts/run_x_search.py \
  "xAI" --params-json '{"from_date":"2026-05-01"}'

The wrapper first tries keyword parameters when possible. If the installed x_search_tool rejects them, it falls back to a single enriched query string.

Workflow

Convert the user's request into a focused X search query.
Run scripts/run_x_search.py through uvx --from hermes-agent python.
Wait patiently; x_search_tool can take 30 seconds or longer.
Inspect the answer for the required fields: link, body text, like count, and author.
If required fields are missing, rerun with wording such as: 各投稿ごとに投稿者名、@handle、いいね数、投稿本文、投稿URLを必ず出力.
Return a concise synthesis plus the per-post evidence. Keep each post's source URL attached to its text.

Final Answer Shape

Prefer this format:

**要約**
<全体傾向を2-4文>

**投稿**
1. <短い見出し>
   - 投稿者: <display name>（@handle）
   - いいね数: <number or 不明>
   - 投稿日: <date/time or 不明>
   - 投稿本文: <text or excerpt + summary if very long>
   - 投稿URL: <url>
   - 外部リンク: <urls or なし>

For long posts or articles, do not paste excessive copyrighted text. Include a short excerpt and summary while keeping the source URL.

Environment Checks

If uvx is missing, tell the user to install or expose uv/uvx in PATH before using this skill. If Hermes authentication fails or returns empty results, check Hermes/X authentication outside the skill.