Explore the fictional Meridian investigation, see what makes a connection meaningful, and create an atlas you can keep.
01 START IN THE CASE
Four things to inspect.
The Meridian File asks who can influence a proposed compute-licensing rule—and what would distinguish patron direction from coordinated or independent agreement. Every actor, amount, document, and quotation is invented teaching material.
Use your browser’s Back action to return to this guide, or open the atlas in a separate tab using your browser’s link menu.
Read the right, not just the grant.
In Atlas, select Cobalt Foundation, then inspect its approval-right relationship with the Licensing policy brief. The agreement gives Cobalt limited review and revision rights over that brief. Compare this with the much broader, disputed allegation about Astera AI.
Follow a source back to its origin.
Open Sources and inspect the direction story and its derivatives. Sentinel, Ledger, and Streamcast trace back to one grant announcement. The repeated claim has a circulation history; repetition does not supply missing evidence of orders.
Let the dates challenge the theory.
Open Explanations and read Shared briefing channel without a central controller. Astera’s June 1 endorsement precedes the recorded July 2 packet. Then inspect the dated relationships in Timeline. This packet cannot explain that initial endorsement.
Find the next discriminating record.
Open Leads. Look for the recipient record early enough to test a shared briefing channel. Compare the expected observation with the alternative result. In Actor games, inspect the assumptions behind possible moves before treating a dependency as usable leverage.
Try Save view after choosing filters or a relationship. The downloaded snapshot preserves presentation state for this case. Load view restores a compatible snapshot; it does not edit the evidence.
02 GIVE YOUR AI THE CRAFT
Install the complete skill.
Download red-threads-v0.1.0.zip. Follow your host’s supported skill installation route and keep the red-threads folder intact, including its references, scripts, assets, and example. The installation guide also covers the maintained src/red-threads source folder and the Codex installer.
Your AI host supplies the model and research tools. Python 3.10+ is needed for the bundled case commands and atlas generation. You can still investigate conversationally without Python.
Uploading a ZIP to an ordinary chat is reference use unless the host actually installs and discovers the skill. Use the host-specific installation guide to choose the right route.
For this local walkthrough, extract the ZIP and open a terminal in the red-threads folder containing SKILL.md. If you use a source clone, open its src/red-threads folder instead. With Python 3.10+ available, run these commands in order. The output goes in the parent directory, outside the skill folder.
A successful result includes "ok": true. Open meridian-atlas.html in the parent folder using a modern browser. The generated file contains the case and works offline; rendering makes no network calls.
The output must be new. If meridian-atlas.html already exists, use ../meridian-atlas-2.html. If the parent folder is not writable, choose a new absolute output path in your own workspace.
If Python or local browser access is unavailable, continue with the public fictional demo or a focused textual map in your AI host. The recovery guide covers missing commands, validation messages, and output conflicts.
04 FOLLOW YOUR OWN THREAD
Bring the material you have.
Supply the document, transcript, question, or existing case you want to investigate. Here is a starting request included with the skill:
A STARTING REQUEST
There may be a story in this transcript. Map the money, roles, resources and decision rights behind it. Show me the most useful geometry, then follow the strongest lead.
Choose a workspace for your case files outside the installed skill. Keep original records, source locators, and material corrections with the investigation. Your case.json is the canonical case; the atlas and exports are derived views.
For an expanding case, use the next leads, coverage gaps, and saved checkpoint to resume. If a source changes, inspect its declared dependencies, review the affected interpretations, and write a new case revision.