Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/xxyoudeadpunkxx/signal-rail/llms.txt

Use this file to discover all available pages before exploring further.

This guide walks you through deploying a Signal Rail instance, opening it with an AI agent, and routing your first material into the correct rails. Follow the steps in order — the entry protocol exists for a reason, and skipping any step creates exactly the undifferentiated material that Signal Rail is designed to prevent.
1

Choose a deployment method

Signal Rail provides two ways to create a deployed instance. Both produce the same result: a copy of all canonical files in your chosen directory.Windows bootstrap (recommended)Run the launcher from the Signal Rail source folder:
init_signal_rail.bat
The launcher prompts for two inputs:
  • Target directory — the folder where the instance will be created. This must be different from the Signal Rail source folder.
  • Host project name (optional) — if provided, the script pre-fills the project name: field in both 01_orientation.txt and 03_master_working.txt and stamps the current date in 03_master_working.txt.
Internally, the launcher calls init_signal_rail.ps1, which copies all canonical files to the target directory, creates the directory if it does not exist, and reports which files were created, overwritten, skipped, or customized.Manual copyCopy the Signal Rail files into the target folder yourself. Manual copy is valid, but the copied instance must preserve the required file names, marker structure, canonical roles, and entry flow. No files may be renamed, merged, or dropped.
2

Open the instance with an AI agent

Open an AI chat session in the context of your deployed instance. Then give the agent these exact instructions in order:
read 00_runtime_entry.txt
Wait for the agent to confirm it has read and understood the runtime entry file. Then:
read 06_ai_to_ai.txt
Reading 06_ai_to_ai.txt activates the Signal Rail protocol. This activation has a specific, bounded effect:
  • It enables the entry protocol and operating rules.
  • It does not bootstrap the instance automatically.
  • It does not classify or write anything automatically.
  • It does not infer the host project on its own.
If AI_TO_AI__DEPLOYED_INSTANCE_SIGNAL_RAIL.txt is present in the folder, the agent will treat the folder as a deployed Signal Rail instance. That marker identifies the folder as a tool — not as the host project.
The deployed instance marker (AI_TO_AI__DEPLOYED_INSTANCE_SIGNAL_RAIL.txt) does not identify the host project. It only says that this folder is a Signal Rail instance. Always close the host project explicitly in the next step — never assume the agent inferred it from the folder name or the marker file.
3

Close the minimum frame

Before the agent may take any canonical action, it must close five things. These are not formalities — they are permission gates. Until each one is closed, the agent is in orientation mode, which is read-only.The five things the agent must close:
  1. Host project — what real project is being governed by this instance
  2. Working object — the specific thing being worked on right now
  3. Active mode — one of: orientation, reference, review, ingest, work, or structure
  4. Source scope — the boundaries of what is in scope for this session
  5. Authority — which source has governing authority for the current task
Orientation mode is read-only. Reading the canonicals does not authorize writing. Understanding a file does not authorize promoting its contents. The agent must explicitly move to a mode that allows writing before any canonical update can occur.
If any of these five items cannot be closed, the agent must stop and ask — not guess, not infer, and not proceed.
4

Complete the minimum read

After the frame is closed, the agent must complete a minimum read before any substantive action. This protects the project from wrong entry even when the frame appears clear.Mandatory reads (always):
FileWhy
01_orientation.txtCloses project identity, perimeter, and reading frame
03_master_working.txtCloses current live state, dominant blocker, and next move
Conditional reads (when relevant):
FileRead when…
02_protocol_freeze.txtIdentity, freeze, or structure may be touched
04_decision_log.txtDecisions may be touched
08_surface_map.txtTechnical surface, entrypoints, sensitive files, or dependencies may be touched
09_handoff_reentry.txtContinuity or restart matters
05_latent_ideas.txtUnresolved live material is under review
A task arriving before minimum read is complete does not authorize skipping minimum read.
5

Fill the two live rails first

If this is a new instance, the canonical files are still template-like. The first thing to do is populate the two rails that govern project identity and current live state. Do not route other material until these two are live.01_orientation.txt — project identityFill this file with the host project’s identity: what it is, why it exists, its core direction, accepted trade-offs, and main boundaries. Write about the host project — not about Signal Rail. Do not put current work, blockers, decisions, or unresolved ideas here. Those belong in other rails.03_master_working.txt — current live stateFill this file with where the project is right now: the current objective, the dominant blocker, and the next sensible move. This is the live pulse of the project. It changes as work changes. It is not an orientation log or a history.Once both files are live, the system has a closed frame and the agent can route additional material accurately.
6

Start routing material

With the frame closed and the two live rails filled, you can begin routing material by nature. Classify the unit first, then assign it to the matching level — not the nearest file.
If the material is…Route to
Current live work — guiding present action03_master_working.txt
A decision that has already won against a real alternative04_decision_log.txt
An open idea — important but still mobile05_latent_ideas.txt
A hard-to-reopen identity constant02_protocol_freeze.txt
Technical topology, entrypoints, or sensitive surfaces08_surface_map.txt
Useful but not active right now98_parking.txt
Closed or historical99_archive.txt
A good sentence is not yet a decision. A strong idea is not yet freeze. Continuity is not the same thing as truth. If the destination is not clear enough, hold the material in 05_latent_ideas.txt until it can be properly classified.

​
After the bootstrap

Once the bootstrap script completes successfully, it prints the following recommended next steps. Follow them in order: 
Recommended next steps:
1. Read 00_runtime_entry.txt first.
2. Read 06_ai_to_ai.txt after entry is valid.
3. Close host project, working object, and mode.
4. Fill 01_orientation.txt and 03_master_working.txt first.
5. Continue from the textual system and use the canonicals by level.
6. Use 97_field_findings.txt only when you want to keep lateral findings readable before routing or discarding them.
These steps are not suggestions — they are the minimum path to a valid governed session. Skipping step 1 or 2 means the agent is operating without an active protocol. Skipping step 3 means the agent has no closed permission boundary. Skipping step 4 means the canonicals are still in an uninitialized state and cannot be treated as live authority surfaces.

Build docs developers (and LLMs) love

Get started for freeTalk to us