Signal OS / Communication Signal Engine

Captured: 2026-04-27 (per GitHub issue zee78900/HookStreet-Business-OS#1)
Status: Design + scaffolding. Markdown + JSON only. No app yet.
Repo decision: Lives in docs/SIGNAL_OS/ of hookstreet-workspace until it earns its way to a separate repo.

Working definition

Capture messy communication. Extract truth, decisions, risk, tone, and next action. Preserve context. Drive execution.

This is not a WhatsApp summarizer. It is a repeatable cognitive workflow for turning messy communication into usable operating intelligence — with structured outputs (decisions, action items, open loops, risks, tone, draft replies, transfer packets) that can be retrieved and reused later.

Why it matters

The deeper goal is an external-brain workflow where scattered communication becomes:

Clear thinking
Decision support
Follow-up execution
Reusable context
Operating memory
Weekly review material

Key insight: Preserve the intelligence of the conversation, not just the content of the conversation.

Connections to existing Hook Street systems

System	How it touches Signal OS
`braindump-router` skill (in `hookstreet-skills/braindump-router/`)	A partial implementation — routes voice dumps into Reminder / Calendar / Task / Follow-up. Signal OS extends this with structured packets, decision logs, transfer packets.
`hookstreet-voice-intake` repo	Existing iPhone Wispr Flow → Gmail [INTAKE-V1] → Apps Script → BOS pipeline. Signal OS reuses this capture mechanism.
EG-CONTEXT-PACK v3 + `reusable-engagement-context-prompt.md`	Eden Gardens consulting context pack is a real-world Signal OS output. The Reusable Prompt template (in `docs/HOW_TO/`) is the engagement-specific variant.
HOW-TO library (`docs/HOW_TO/`)	Each HOW-TO doc was a Signal OS extraction event — voice/text → durable structured doc.
BOS Sheets v1 + v2	Final destination for action items, obligations, calendar items. Signal OS feeds them.

The repeatable loop

1. CAPTURE     → low-friction (voice / WhatsApp export / Gmail / typed)
2. NORMALIZE   → wrap input in standard envelope (see normalize-format.md)
3. EXTRACT     → run "Extract Signal" prompt → structured packet (JSON)
4. DECIDE      → human review of packet — confirm decisions, owners, dates
5. ROUTE       → action items → Reminders / Calendar / monday.com / BOS
6. PRESERVE    → archive packet for retrieval (search-by-person, by-date, by-topic)
7. REVIEW      → weekly synthesis (see templates/weekly-review.md)

Folder structure (this scaffold)

docs/SIGNAL_OS/
├── README.md                       (this file)
├── normalize-format.md             (the wrapping envelope for any input)
├── prompts/
│   ├── extract-signal.md           (raw input → structured packet)
│   └── transfer-packet.md          (preserve full context for AI handoff)
├── schemas/
│   └── conversation-packet.json    (JSON schema — the structured output)
└── templates/
    └── weekly-review.md            (Sunday/Monday synthesis template)

Doctrine

Execution is king. Every output must be actionable.
Low-friction beats elegant. Capture in whatever channel is open.
Markdown + JSON before apps. No SaaS, no dashboard, no DB until the workflow proves itself.
Separate facts from inference. Tag every claim. Flag uncertainty.
Don't invent. If a name/date/dollar isn't in the source, it doesn't exist.
Preserve source of truth. Original transcript stays alongside the extracted packet.
Batch beats scatter. Process 10 messages once a day, not one at a time all day.
Compact logs beat bloated archives. Every packet should be retrievable in <30 sec.

When to use vs. skip

USE Signal OS when:
- Long WhatsApp / SMS thread you want to extract decisions from
- Voice dump that mixes 5+ topics
- Meeting transcript needing action capture
- Engagement context handoff to a future AI session
- Weekly review of "what did I commit to and to whom?"

SKIP Signal OS when:
- Single-message reply (just answer it)
- Already-structured email (just file it)
- Real-time chat (use braindump-router skill instead)

What's NOT in this scaffold yet

Automation. (Manual run-the-prompt-yourself for now.)
Routing into Reminders / Calendar / monday.com. (Manual paste for now.)
Full-text search across packets. (Use grep until it hurts.)
Web dashboard. (Maybe Phase 2 in command-center once the workflow is real.)

Naming options

Signal OS (current working name)
Hook Street Signal Engine
Communication Signal Engine
Conversation Intelligence Loop

Pick one before going public.

Next step (when ready)

Test the loop manually on 3 real inputs (one WhatsApp export, one voice transcript, one email thread). Run prompts/extract-signal.md against each. Save outputs as archive/2026-04-27/<source>.packet.json.
Iterate the prompts based on what came out wrong.
Once the loop produces value 5+ times, automate the EXTRACT step (likely an Apps Script that takes a Gmail label Signal-Inbox and writes packets to a sheet).
Once retrieval starts to hurt (>30 sec to find an old packet), build the search layer.

Source

GitHub issue zee78900/HookStreet-Business-OS#1 (full design captured 2026-04-27)
Cross-reference: hookstreet-skills/braindump-router/SKILL.md
Cross-reference: docs/HOW_TO/reusable-engagement-context-prompt.md
Cross-reference: docs/HOW_TO/whatsapp-chat-extraction.md