x-read

Use the obsidian-second-brain skill. Execute /x-read [url]:

1. Resolve the URL from the user's argument. If no URL was given, ask: "Which X post URL?" Accept any URL containing x.com/ or twitter.com/.

2. Run the script from the skill root (its absolute path was given at session start as Skill root; substitute it for SKILL_ROOT):

   uv run --directory "SKILL_ROOT" -m scripts.research.x_read "<url>"
   

3. The script prints a structured analysis (ORIGINAL POST, THREAD, TL;DR, KEY CLAIMS, REPLY SENTIMENT, NOTABLE COUNTER-ARGUMENTS, VOICES TO WATCH) and a one-line cost summary on stderr. Show the analysis to the user verbatim - don't paraphrase or summarize.

4. Default save behavior: chat only. Do NOT save the analysis to the vault automatically. The user must ask explicitly ("save this", "save to vault", "/obsidian-save") for it to be archived.

5. If the user asks to save: write an AI-first note to Research/X-reads/YYYY-MM-DD - <slug>.md in the vault, following the AI-first vault rule (Section 0 of _CLAUDE.md):

   • Frontmatter: date, time, type: x-read, ai-first: true, post-url, post-author (if known), key-claims (list), tags, related-people (wikilinks for any @ handles that map to known people in the vault), cost-usd
   • Body starts with a ## For future agent preamble section (2-3 sentences summarizing what this post is about and why it was saved)
   • Then the full structured analysis from the script

6. Plain English triggers that route to this command: "read this tweet", "read this X post", "what's in this tweet", "analyze this X link" - when followed by a URL.

7. If the script fails with a clear error (missing key, network down), surface the error message verbatim. The script handles retry on transient errors automatically.

AI-first rule: Every note created or updated by this command MUST follow references/ai-first-rules.md - ## For future agent preamble, rich frontmatter (type, date, tags, ai-first: true, plus type-specific fields), recency markers per external claim, mandatory [[wikilinks]] for every person/project/concept referenced, sources preserved verbatim with URLs inline, and confidence levels where applicable. If that path does not resolve from your working directory, search upward for it; if you still cannot read it, say so before writing rather than producing a note that silently skips the rule. The vault is for future agent retrieval - not human reading.

Anti-fabrication: Search exhaustively before claiming any note, person, or file is absent - false absence is the most common failure mode - and never invent facts, entities, or dates (mark unknowns as TBD). See the anti-fabrication and search-completeness hard rules in references/ai-first-rules.md.