sills

Sills

Engineering intelligence skills for AI coding agents.

Sills provides report-only Audit Skills that discover a software project, build reusable project knowledge, gather evidence once, and produce consistent findings, coverage, trends, risks, and remediation handoffs.

Supported out of the box

Support depth varies by platform. Sills distinguishes detected, normalized, collected, runtime-capable, and specialist-aware support instead of treating every integration as equally complete. See the support matrix.

How Sills works

Project
  ↓
Discovery and built-in platform collectors
  ↓
Shared project knowledge and evidence
  ↓
Selected Audit Skills at Quick, Standard or Deep depth
  ↓
JSON, Markdown and HTML reports
  ↓
Verification and remediation handoff

The umbrella skill does not run every audit by default. For a generic request, it discovers the project, recommends relevant audits, asks for an explicit audit selection and depth, gathers reusable evidence once, and coordinates only the selected Audit Skills.

Audit depth

Mode and depth are independent. A run may be source + deep, full + standard, or changed + quick. Reports record both the requested depth and the coverage actually achieved. See Audit depth.

Why Sills is different

• Report-only by design. Audits do not modify the product outside the selected audit directory.
• Evidence before judgement. Findings reference source, runtime, logs, screenshots, traces, configurations, or explicit limitations.
• Shared project knowledge. Routes, services, platforms, deployments, auth, data, tests, and other facts are collected once and reused.
• Stable findings. Deterministic identities support verification and trend tracking across runs.
• Honest coverage. Missing roles, runtime targets, simulators, tools, or environments remain visible.
• Agent-portable. Skills use the shared Agent Skills format for Claude Code and OpenAI Codex.

Audit Skills

Auditing is the first application of the Sills knowledge and evidence model. Future Skill categories may consume the same contracts for safe fixes, planning, migrations, documentation, and project understanding.

Install

npx sills-audit install

This installs the $sills shortcut, the $sills-audit umbrella skill, and all specialists into both .agents/skills/ and .claude/skills/ in the current project. Use --codex, --claude, or --global to narrow the destination.

npx sills-audit install --codex
npx sills-audit install --claude --global
npx sills-audit-agent-readiness install

Run

$sills-audit Audit this project.
$sills audit api

A generic request prompts for both the relevant Audit Skills and one of three depths: Quick, Standard, or Deep. To request every applicable Audit Skill explicitly:

$sills-audit Do a full standard audit of this project.

The default output is a dated folder:

audit/2026-07-15/
├── index.md
├── summary.md
├── report.md
├── report.json
├── report.html
├── manifest.json
├── handoff.md
├── coverage.json
├── project-knowledge.json
├── reports/
├── findings/
├── evidence/
└── raw/

Documentation

• Support matrix
• Audit depth
• Terminology
• Discovery and platform support
• Shared project knowledge
• Platform and knowledge roadmap
• Remediation engine plan

Development

npm install
npm test
npm run validate
npm run pack:check

Public website content lives in web/docs.

Status

Sills is an open-source beta. The repository provides fourteen specialist Audit Skills, a selective umbrella orchestrator, built-in platform discovery, shared knowledge and evidence contracts, report schemas, audit scaffolding, fixtures, CI, Cloudflare deployment, and npm release automation. Audit quality still depends on the capabilities available to the invoking agent and environment, so every report must state its coverage and limitations.

License

MIT