| name | x-thread-context-capture |
| description | Build a single-file research context pack around one X/Twitter post by browsing `x.com` in Chrome and extracting the root post, author follow-ups, key replies, key participants, and important internal/external references. Save the result into `cwd/x-thread-contexts/` using a structured filename. Use when the user shares an `x.com` or `twitter.com` status URL and wants the complete thread, author replies, key comments, or a model-ready context package for later deep research, analysis, or agent workflows. |
X Thread Context Capture
Produce a high-density context package for one X post, not an opinion piece. Preserve original text, preserve source boundaries, and give the next model enough first-hand material to continue the research without reopening X.
The output contract is strict: deliver one Markdown file that stands on its own, saving it under cwd/x-thread-contexts/ with the default filename structure x-thread-context-<author_handle>-<root_status_id>.md.
Output Contract
- Deliver a single Markdown file under
cwd/x-thread-contexts/, named x-thread-context-<author_handle>-<root_status_id>.md by default.
- Preserve raw text over summary whenever possible.
- Separate author content, comments, X-internal referenced content, and third-party links.
- Record visible metadata for important items: author, handle, timestamp, and visible interaction counts when available.
- Treat counts as unstable page snapshots, not durable facts.
- Add a short coverage note when the visible page suggests the thread may be incomplete or when no more author replies can be found.
- Use the root-post author handle without
@, lowercase it, and use the root status id from the URL. If the handle is temporarily unavailable, use unknown-author and rename once confirmed.
- Create
x-thread-contexts/ under the current working directory if it does not already exist.
Read references/final-template.md right before drafting the deliverable.
Workflow
- Open the root status URL on
x.com in Chrome DevTools MCP and read it from snapshots first.
- Capture the root post text, author, canonical URL, timestamp, and visible interaction signals.
- Prefer author context before broad comment harvesting.
- Open the author's thread view at
https://x.com/<handle>/thread/<root_status_id> when it exists.
- Run an X site search for
from:<handle> conversation_id:<root_status_id> in Latest to find author replies that do not appear cleanly in the default conversation view.
- Open high-value author replies individually when needed so you can capture both the upstream comment and the author's reply with exact text.
- Collect a selective set of high-value comments after author coverage is solid.
- Capture X-internal quoted or linked posts directly inside the deliverable file; keep third-party links as URLs plus visible preview text unless the user explicitly asks for more.
- Stop when the root post, author follow-ups, key comments, and key references form a clear research context pack and additional scrolling is mostly repetitive.
- Save the deliverable into
cwd/x-thread-contexts/.
- After saving the deliverable, close the X tabs opened for this task so the browser does not accumulate research tabs over time.
Required Rules
- Use Chrome browsing behavior as the primary source of truth. Stay within ordinary browsing, scrolling, clicking, waiting, and reading.
- Do not rely on X APIs, unofficial APIs, network interception, or DOM scraping as the main acquisition path.
- Prefer X snapshots and individual post pages over third-party mirrors for post reconstruction.
- If login walls, rate limits, deleted posts, or region blocks prevent access, state exactly what is blocked and what was still recovered.
- Do not merge different source types together. Keep root post, author replies, comments, quoted posts, and external references distinct.
- Do not let system explanation dominate the deliverable. The Markdown file is a context package, not a commentary essay.
- Close only tabs opened for the task. Do not close unrelated pages the user may already have open.
- Do not scatter result files across the cwd root. Use the dedicated output directory unless the user explicitly asks for a different location.
Selection Rules
- Cover all author-originated material first: root post, self-thread entries, clarifications, and direct replies that materially change understanding.
- Prefer comments from high-signal participants: verified/high-profile accounts, comments with strong interaction signals, comments the author replied to, and comments that introduce a new frame, example, objection, or resource.
- Skip low-information replies such as applause, pure emotion, repetition, and obvious topic drift unless they matter because the author engaged with them.
- When in doubt about comment selection or stopping conditions, read references/selection-rubric.md.
Writing Rules For The Deliverable
- Use short, stable section headings.
- Put the exact URL on every important item.
- Use absolute dates or the exact timestamp as shown on X when available.
- Quote the raw post text in fenced
text blocks when that improves fidelity.
- Add a short "coverage boundary" section at the end if you had to infer completeness from what X exposed.
- Explicitly say when you searched for more author replies and did not find any further visible results.
- Save to
cwd/x-thread-contexts/x-thread-context-<author_handle>-<root_status_id>.md unless the user explicitly requests a different location or filename.
Useful Tactics
- On long posts, use the expanded post view and individual status pages to avoid truncated text.
- On author replies, the individual status page is often the easiest place to recover both the upstream comment and the response cleanly.
- The search pattern
from:<handle> conversation_id:<root_status_id> is the fastest way to discover author replies scattered through the conversation.
- The
/thread/<root_status_id> route is the fastest way to check whether the author published a continuous self-thread.
References