get core

free · local · open

Install once.
Activate.
Build better.

Specability Core is the free local coding-agent harness for the coding domain. One static binary, no cloud dependency, and no independent upload to SpecabilityAI. Enabled host hooks send bounded orientation to your selected AI host and its model provider. Install it, activate for your AI coding agent, and start working.

setup · step 1

One command. Any platform.

Specability Core ships as a single static binary. Pick whichever channel fits your workflow — they all install the same thing. After installing, continue to Step 2 to activate Core for your agent. Installation alone does not connect it.

npm
npm install -g specability
Homebrew (macOS / Linux)
brew tap SpecabilityAI/tap && brew install specability
shell script (macOS / Linux)
curl -fsSL https://raw.githubusercontent.com/SpecabilityAI/specability-core/main/install.sh | sh
PowerShell (Windows)
irm https://raw.githubusercontent.com/SpecabilityAI/specability-core/main/install.ps1 | iex

setup · step 2

Tell your agent where the harness is.

Core connects through a single session-start hook — hook-only. No skill files, agent definitions, or spec files are written into your host directories. Without this step, Core is installed but not active in your agent sessions.

  1. one

    Register the hook

    Pick the agent you use. The hook fires at session start and injects orientation context automatically.

    specability install hook --host claude --scope global   # Claude Code
specability install hook --host codex  --scope global   # Codex CLI
specability install hook --host gemini --scope global   # Gemini CLI
  2. two

    Open a new session

    Start a fresh agent session — the hook fires on session start only, so existing sessions do not pick it up. On first run, your agent host may prompt you to allow the hook to execute: approve it, that is Core connecting. The hook then supplies bounded project orientation and methodology pointers to that host; this becomes context for its configured model provider.

  3. three

    Scope to a project (optional)

    Use --scope project instead of --scope global to activate Core for a single project only. Useful when you want the harness in some repos but not all.

    cd your-project
specability install hook --host claude --scope project

setup · step 3

Confirm everything works.

specability doctor

where core fits

Three layers. One enforces.

Skill frameworks shape how your agent works. Spec tools align your agent on what to build. Core is the enforcement layer — the one that determines whether your agent actually did it.

  1. skill frameworks

    e.g. Superpowers

    Deliver methodology as skill files injected at session start. Shape how the agent works — its process, roles, and thinking — before any work begins.

    The agent reads the skill. Whether it follows through — and whether the work is genuinely done — is not enforced at runtime.

  2. spec tools

    e.g. OpenSpec

    Define what should be built before coding starts. Align human and agent on requirements via structured spec artifacts.

    The spec says what to build. Whether the agent built it — and can prove it — is outside the spec layer.

  3. process harnesses

    Specability Core

    Enforce the process at runtime. Gate completion on evidence. Record every task, decision, check, and result in a durable, searchable trail.

your first session

Open a session. The harness takes it from there.

After activation, every new agent session begins with structured orientation from Core. The agent knows the project, the methodology, and what it needs to verify before calling work done. You describe the task — the harness holds the process.

  1. one

    Structured context at session start

    The hook fires once. Your agent receives the registered project map, methodology pointers, and safe runtime metadata automatically — without absolute local paths or memory claim bodies. Explicit task recovery is included only when a task ID is supplied.

  2. two

    Your agent discovers and documents the codebase

    When a stable module needs durable orientation, the agent creates and maintains a README describing its role, ownership, touchpoints, and change guidance, then registers its exact path. The CLI consumes only explicit registrations and performs mechanical checks; semantic reconciliation stays with the agent.

  3. three

    Your agent works under constraint

    Defined roles. Mandatory evidence checks. Boundary compliance enforced by the harness. The agent cannot mark work done by self-report alone — the evidence must be on record, and the harness reads the record.

  4. four

    You get the record without asking for it

    Every task, decision, check, and result is written to a durable, searchable record. When a decision is questioned later, the answer — and the rule it followed — is already on file.

source

Open builds. Public changelogs.

Every release ships binaries, SHA-256 checksums, and release notes. Code signing is not yet provided. The public repository hosts install scripts, the npm wrapper, and the Homebrew formula. File issues, read changelogs, and verify builds — all in the open.

GitHub → SpecabilityAI/specability-core
← how the harness worksrequest an engagement →