# Add to your Claude Code skills
git clone https://github.com/mattbaconz/signal
README.md
SIGNAL is a brutalist compression layer for agentic workflows: minified .min.md skills, a small symbol vocabulary (→∅Δ![n]), and checkpoints instead of pasting entire threads. Dense beats polite when the meter is running.
Latest changes:CHANGELOG.md (v0.3.1 patch notes, v0.3.0 “Shrinking Session”, and earlier). Contributing / min skills:CONTRIBUTING.md.
Demo
Heuristic token estimates use ceil(characters / 4) — useful for comparing shapes, not billed API tokens. Full tables: Benchmark. Reproduce: scripts/benchmark.ps1.
Video/GIF: A short install or benchmark run GIF is planned; see assets/README.md to record assets/signal-demo.gif and swap it into this section later.
Before / after
| 👤 Verbose agent | 🌐 SIGNAL |
| --- | --- |
| “I think the problem might be in around line 47…” | · null ref · guard — in the scripted benchmark. |
| Paste 10 turns of chat + tool noise into context. | : stack, progress, next step — transcript stays out of the window. |
| One giant tree + references forever. | for humans, for the agent — smaller on the seven main pairs (). |
There is no legacy top-level signal/ directory in this repo—ignore older docs that referred to it. This table is what you actually open after cloning.
| Location | What it is | You need it if… |
| --- | --- | --- |
| skills/ | Canonical skill specs: *.md (readable) + *.min.md (dense) | You install via npx skills add or copy skills into an agent |
| gemini-signal/, claude-signal/ | Mirrored host extension layouts (SKILL.md per tool) | You ship the Gemini CLI or Claude Code plugin from this tree |
| references/ | Shared refs (symbols, Karpathy norms, benchmarks, checkpoint notes) | You cite norms or symbols |
| templates/ | Snippets to merge into a project’s GEMINI / CLAUDE files | You integrate SIGNAL into an app repo |
| scripts/ | shrink.ps1, sync-integration-packages.ps1, verify.ps1, benchmark.ps1 | You contribute or verify locally (CONTRIBUTING.md) |
Why SIGNAL
Long prompts and hedging eat context. SIGNAL standardizes how you shrink: symbols instead of paragraphs, .signal_state.md for durable state, signal-diff / signal-search for summarized context instead of raw dumps.
flowchart LR
subgraph inputs [Verbose]
V1[Long replies]
V2[Full thread history]
V3[Hedging and filler]
end
subgraph signal [SIGNAL]
S[Symbol grammar]
C[CKPT summaries]
M[Minified skills]
end
subgraph outputs [Dense]
D1[Terse pointers]
D2[Checkpoint atom]
D3[~87% smaller payloads]
end
V1 --> S
V2 --> C
V3 --> S
S --> D1
C --> D2
M --> D3
Tiers
Use /signal, /signal2, or /signal3.
| Tier | You get | Rough habit savings |
| --- | --- | --- |
| S1 | Symbols, no preamble, no hedge, terse | ~35% |
| S2 | S1 + BOOT, aliases, delta-friendly turns | another ~20% on top |
| S3 | S2 + auto-checkpoint every 5 turns | long sessions stay bounded |
flowchart TB
subgraph protocol [Protocol layer]
P["signal.min.md"]
CORE["signal-core.min.md"]
end
subgraph workflow [Workflow skills]
CKPT["signal-ckpt"]
GIT["commit / push / pr"]
REV["signal-review"]
end
subgraph context [Context tools]
DIFF["signal-diff"]
SRCH["signal-search"]
ST["signal-state"]
end
subgraph persist [Persistence]
STATE[".signal_state.md"]
end
P --> CORE
P --> CKPT
P --> GIT
P --> REV
P --> DIFF
P --> SRCH
ST --> STATE
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
