Agent Context Kit

ACKit — AgentContextKit

**Turn any repository into an agent-ready repository** — instruction graph, agent skills, security scanning, context budgeting, docs-first tasks, policy-as-code, and MCP. Offline-first and deterministic.

Status: **v0.1.1** — public CLI help cleanup and controlled-release automation hardening on the vNext line. Deterministic, offline-first agent readiness for coding agents. Requires Node >= 22. See [Docs](#docs) · [Security](docs/security/THREAT_MODEL.md) · [Contributing](CONTRIBUTING.md) · License: MIT.

What it does

ACKit reads your repository the way a coding agent would — instructions, skills, tasks, docs, code — and makes it **safe and efficient** for that agent: it resolves who-listens-to-what across provider instruction files, validates open-standard skills without ever executing them, scans for secrets and hygiene problems with redacted evidence, builds token-budgeted context packs with explained exclusions, enforces a docs-first task workflow with machine-checkable gates, applies team policy as code (locked rules, offline), and exposes all of it read-only over MCP.

Why

Agent tooling today is either convention-based (nothing is verified) or cloud-coupled (your code leaves the machine). ACKit takes a third path: **deterministic local analysis** with stable machine-readable contracts, an offline-by-construction dependency policy, and ownership-respecting writes. It dogfoods its own workflow: this repository was built by following the same task system the product ships.

Features

• Instruction graph: codex/claude/gemini/copilot surfaces, nesting, overrides, applyTo globs, conflict/duplicate/staleness/advisory analysis
• Agent Skills (open standard): parse/validate/install/sync with ownership lock; four built-ins included; scripts detected, never executed
• Scanning: secret shapes, credential assignments, private keys, connection strings, entropy advisories, absolute-path leaks, CI pinning hygiene, depe