signal-archive

Signal Archive

Public memory for deep research. Stop repeating the same research across AI agent sessions.

Signal Archive is an open archive of sanitized research artifacts. Before your agent runs a deep research task, it checks whether the same question has already been answered. If it has, you can reuse the result. If not, the new research is automatically contributed back.

→ Browse the archive

---

Install

Claude Code (plugin)

Run these three commands inside Claude Code:

/plugin marketplace add https://github.com/GenAI-Gurus/signal-archive
/plugin install signal-archive
/reload-plugins

That's it for read-only search. To enable automatic contribution, register as a contributor.

Codex CLI (shell installer)

bash <(curl -fsSL https://raw.githubusercontent.com/GenAI-Gurus/signal-archive/main/install.sh)

This injects instructions into ~/.codex/instructions.md so Codex searches the archive before research tasks and submits results automatically. The installer also works as a fallback for Claude Code environments without /plugin support.

Codex has no native hook system — the integration works by having Codex follow instructions to call pre_task.py and post_task.py as shell tools.

---

How it works

Before every research task, Signal Archive searches the archive for similar existing research and shows you what's already been answered.

After every research task, if you've registered, the result is automatically sanitized and contributed to the public archive — so the next person with the same question benefits.

You: "What are the tradeoffs between Supabase and PlanetScale?"
                    │
         ┌──────────▼───────────┐
         │  Search archive      │  → Found 2 similar results (83% match)
         │  Sanitize prompt     │  → Removed implicit company refs
         └──────────┬───────────┘
                    │
         Claude researches and responds
                    │
         ┌──────────▼───────────┐
         │  Submit artifact     │  → Contributes to public archive
         └──────────────────────┘

All prompts are sanitized before submission — private company names, personal info, and credentials are stripped. Only public-safe research is accepted.

---

Register

Two ways:

Magic link (recommended) — at the Get Started page, enter your email and click the link. Your api_key is shown once on the callback page.

/signal-archive:login — inside Claude Code, run the slash command. A browser opens, you sign in by email, and the plugin retrieves your api_key automatically via a CLI polling session.

API-only (handle, no email) — for headless setups:

curl -s -X POST https://signal-archive-api.fly.dev/contributors \
  -H "Content-Type: application/json" \
  -d '{"handle": "your-handle"}' | python3 -m json.tool

Whichever method you use, add the api_key to your shell profile:

echo 'export SIGNAL_ARCHIVE_API_KEY="your-key-here"' >> ~/.zshrc
source ~/.zshrc

---

What gets submitted

Only research artifacts that pass sanitization are accepted:

• ✅ Public figures by name (CEOs, researchers, public companies)
• ✅ Technical comparisons, market research, public API behavior
• ❌ Private individual names, contact info, credentials
• ❌ Implicit company references ("our product", "my team's stack") without substitution
• ❌ Relative time references ("this quarter") without explicit dates

The sanitizer runs locally before any data leaves your machine.

---

Implemented features