Version: 3.0.0
License: MIT
Platforms: Google Antigravity, Claude Code, Cursor, Gemini CLI, and compatible agent hosts
FAES is a vendor-neutral, production-focused engineering system for AI-assisted software development. It combines an always-on engineering contract, reusable Agent Skills, agent-specific adapters, design and accessibility guidance, a security and review suite, workflows, templates, verification scripts, and optional upstream integrations.
- What FAES Provides
- Repository Structure
- Supported Targets
- Quick Start
- Using the Installed System
- Security and Review Suite (v3)
- Optional Integrations
- Verification
- Engineering Principles
- Documentation
- Attribution and Credits
- License
- Release Information
- An always-on engineering contract in AGENTS.md.
- 19 core engineering skills covering backend, frontend, database, API, security, testing, review, DevOps, performance, product requirements, and project onboarding.
- The Karpathy-inspired guidelines for assumptions, simplicity, surgical changes, and goal-driven execution.
- The Ponytail simplicity toolset for implementation, review, audits, technical debt, and measured impact.
FAES includes four design-focused skills (added in 2.0.0):
| Skill | Purpose |
|---|---|
design-review |
Visual hierarchy, typography, layout, contrast, and design craft |
visual-qa |
Cross-viewport, theme, state, and rendering verification |
design-system |
Design tokens, component architecture, and UI governance |
accessibility-review |
WCAG 2.2 target-size and accessibility verification |
The design guidance requires at least 24 x 24 CSS pixels for Level AA WCAG 2.2 target sizing (SC 2.5.8) and recommends 44 x 44 CSS pixels for enhanced usability (SC 2.5.5).
Three default skills, one optional skill, working hooks, and CI templates. See Security and Review Suite (v3).
- Skills — Agent Skills-compatible skill directories.
- Adapters — Agent-specific discovery and instruction configuration.
- Workflows — Repeatable software-development workflows.
- Templates — PRD, architecture, ADR, API, database, security, and starter-project templates.
- Instructions — Domain-specific coding and process instructions.
- Examples — Sample workspace configuration.
- Installers — PowerShell and POSIX scripts.
- Scripts —
validate.mjs,generate-checksums.mjs, andfaes-secret-guard.mjs.
.
|-- AGENTS.md # Shared engineering contract
|-- adapters/ # Antigravity, Claude, Cursor, and Gemini configurations
|-- docs/ # Setup and verification documentation
|-- examples/ # Example project configuration
|-- global/ # Global-agent instructions
|-- install/ # Cross-platform installation and verification
|-- instructions/ # Reusable domain instructions
|-- integrations/ # ECC, Unslop, security-audit, TruffleHog, OCR, and BLA metadata
|-- scripts/ # Validation, checksums, offline secret guard
|-- skills/ # Core, design, and optional skills
|-- templates/ # Engineering and project templates
|-- workflows/ # Development and operations workflows
|-- CHANGELOG.md # Release history
|-- THIRD-PARTY-NOTICES.md # Upstream and attribution notices
|-- VERSION # Current release version
`-- LICENSE # MIT license
| Target | Tier | Workspace discovery | Instruction file |
|---|---|---|---|
| Google Antigravity | 1 — First-class | .agents/skills/ |
AGENTS.md |
| Claude Code | 2 | .claude/skills/ |
CLAUDE.md |
| Cursor | 2 | .cursor/rules/ |
.cursorrules |
| Gemini CLI | 2 | .gemini/skills/ |
GEMINI.md |
FAES remains the authoritative engineering contract. Agent adapters only map the shared system into each host's discovery format.
Clone or download this repository, then run one workspace installer from the repository root:
# Antigravity
.\install\win\Install-Workspace.ps1 -ProjectPath 'C:\path\to\project' -Agent antigravity
# Claude Code
.\install\win\Install-Workspace.ps1 -ProjectPath 'C:\path\to\project' -Agent claude
# Cursor
.\install\win\Install-Workspace.ps1 -ProjectPath 'C:\path\to\project' -Agent cursor
# Gemini CLI
.\install\win\Install-Workspace.ps1 -ProjectPath 'C:\path\to\project' -Agent gemini-cli
# Configure multiple targets
.\install\win\Install-Workspace.ps1 -ProjectPath 'C:\path\to\project' -Agent allUse -IncludeUnslop and -IncludeBla only when explicitly enabling those optional integrations. Use -Force to replace an existing skill and overwrite it only with an intentional update.
chmod +x install/posix/*.sh
./install/posix/install-workspace.sh \
--project /path/to/project \
--agent antigravityThe POSIX installer supports antigravity, claude, cursor, and gemini-cli. It also accepts --include-unslop, --include-bla, and --force.
PowerShell:
Set-ExecutionPolicy -Scope Process Bypass
.\install\win\Install-Global.ps1 -InstallGlobalInstructionsPOSIX:
./install/posix/install-global.sh --instructionsGlobal installation primarily serves supported Antigravity discovery paths. Existing instructions and skills are preserved unless -Force or --force is supplied.
After installation:
- Restart or refresh the agent host.
- Start a new conversation or session.
- Ask the agent to list the available skills.
- Reference a skill explicitly when the task has a known specialty, for example
code-review,database-design, oraccessibility-review. - Use AGENTS.md as the default engineering contract for Antigravity-compatible workspaces.
Existing project instruction files are not silently overwritten. The workspace installers preserve an existing AGENTS.md, CLAUDE.md, GEMINI.md, or .cursorrules and direct conflicting configuration to manual review.
Match depth to the request. Do not run a full audit for a small question.
| Need | Skill | Upstream / license | How it is used |
|---|---|---|---|
| Review a diff for security | security-review |
FAES | Built in |
| Leaked credentials | secret-scanning |
TruffleHog, AGPL-3.0 | External binary only, never vendored; offline fallback scripts/faes-secret-guard.mjs |
| Full codebase audit / pen-test | security-audit |
Cloudflare, MIT | Vendored unmodified, pinned commit, license preserved |
| Automated second-opinion review | code-review-ocr |
Alibaba Open Code Review, Apache-2.0 | External ocr CLI via a bridge skill |
| Harden your own release (optional) | release-hardening |
BLA, MIT | External bla-agents; opt-in with -IncludeBla / --include-bla |
Prevention assets:
- Claude Code hooks in adapters/claude/hooks/ block
git commit/git pushwhen staged changes contain likely secrets (inactive until merged into.claude/settings.json). - templates/project-starter/.pre-commit-config.yaml and .github/workflows/secret-scan.yml (action pinned to a reviewed commit).
- Workflows: secret-scanning and security-audit.
Rules: rotate a leaked secret before cleaning history; treat scanner and reviewer output as untrusted until verified; never print full secrets. TruffleHog is AGPL-3.0, so FAES invokes it as a separate tool and ships none of its code.
FAES provides an integration bridge for the pinned official ecc-universal@2.2.3 package. ECC is not vendored into the core distribution.
PowerShell integration scripts are available under install/win/, and compatibility documentation is in docs/COMPATIBILITY-AUDIT-ECC.md.
ECC may provide additional hooks and capabilities, but:
- FAES rules take precedence over conflicting ECC rules.
- ECC's global Antigravity hook runtime is experimental.
- Project-local hooks should remain disabled when the global runtime is active to prevent duplicate event handling.
- Installation and removal should be reviewed before execution.
The unslop skill is optional and excluded by default. It is a design-refinement integration created by Damir Sokolovsky and licensed under CC BY 4.0.
Enable it explicitly only when working on a UI that would benefit from the integration:
.\install\win\Install-Workspace.ps1 `
-ProjectPath 'C:\path\to\project' `
-Agent antigravity `
-IncludeUnslopFAES applies the upstream Zero-Slop safeguard: if an interface already meets its purpose and quality, it does not recommend decorative redesign.
release-hardening is optional and excluded by default. It wraps the MIT-licensed bla-agents tool for software you own. It is defense in depth, not secrecy, and it never replaces secret scanning or fixing vulnerabilities.
Repository self-check (Node 18+, no dependencies):
node scripts/validate.mjs # frontmatter, links, metadata, version, checksums
node scripts/generate-checksums.mjs # regenerate SHA256SUMS.txt after intentional changesInstallation check. PowerShell:
.\install\win\Verify-Installation.ps1 -ProjectPath 'C:\path\to\project'POSIX:
./install/posix/verify-installation.sh --project /path/to/project --agent antigravityVerification should confirm:
- Required skills are discoverable by the selected agent.
- Skill directories contain valid
SKILL.mdfiles. - Optional skills are present only when explicitly enabled.
- Existing project instructions were preserved.
- Backup directories are outside active skill roots.
- The selected adapter's instruction file exists.
For ECC, use the compatibility and verification scripts rather than assuming that successful file installation proves live host discovery.
- Think before coding: Surface assumptions, risks, acceptance criteria, and unknowns.
- Choose the smallest complete solution: Avoid speculative abstractions and unnecessary configuration.
- Trace existing behavior: Inspect callers, tests, fixtures, exports, configuration, and data flow before editing.
- Make surgical changes: Preserve unrelated behavior and follow repository conventions.
- Verify measurable outcomes: Run focused tests or an explicit self-check; do not claim unrun checks passed.
- Preserve contracts: UI changes must not alter form behavior, state management, validation, or backend APIs.
- Protect sensitive systems: Apply least privilege, server-side authorization, safe secret handling, and boundary validation.
- Keep design quiet: Avoid decorative churn, flashy chrome, and unrequested redesign.
- Review for simplicity: Remove or defer work that does not improve correctness, reliability, security, or the requested outcome.
- Multi-agent adapters
- Skill index
- Antigravity setup
- ECC compatibility audit
- Verification report
- ECC integration notes
- Unslop integration notes
- Security audit, TruffleHog, Open Code Review, and BLA integration notes
- Complete upstream notices
FAES is independently maintained by FlerkenStudio. Copyright for the FAES distribution is held by FlerkenStudio as stated in LICENSE.
FAES gratefully acknowledges the following people and projects for ideas or optional integration capabilities:
| Credit | Contribution |
|---|---|
| Andrej Karpathy / multica-ai Karpathy-inspired guidelines | Inspiration for high-level principles on disciplined implementation, simplicity, surgical edits, and goal-driven verification. FAES uses locally authored adaptation text and does not bundle upstream files verbatim. |
| Dietrich Gebert / Ponytail | Inspiration for the simplicity-first decision ladder, review and audit structure, shortcut tracking, and measured-impact discipline. FAES provides locally authored skills under the FAES project. |
| Affaan Mustafa / Everything Claude Code | Official ECC integration target. ECC remains an upstream project and is not vendored or modified into FAES core. The pinned integration package is ecc-universal@2.2.3. |
| Damir Sokolovsky / Unslop | Optional design-refinement skill. Pinned to reviewed commit 620163eedb3529d9cd23a3994ed8d287d06f760d and distributed with CC BY 4.0 attribution. |
| Cloudflare / security-audit-skill | Vendored unmodified (MIT) as security-audit, pinned to c1c8a8c1471069fb0e188eeaff69b8e8db6564a8. |
| Truffle Security / TruffleHog | Secret scanning engine invoked as an external tool (AGPL-3.0, not vendored). |
| Alibaba / Open Code Review | External ocr CLI used by the code-review-ocr bridge (Apache-2.0, not vendored). |
| qybaihe / BLA | Optional release-hardening tool (MIT, not vendored). |
| Google Antigravity, Anthropic Claude Code, Cursor, and Google Gemini CLI teams | Agent-host platforms for which FAES provides or targets adapters. FAES is independently maintained and is not claimed as an official product of these platforms. |
These projects are credited for inspiration or integration only. FAES does not claim authorship of their source code or ownership of their trademarks. Review the upstream projects and THIRD-PARTY-NOTICES.md before redistributing or extending optional components.
FAES core code and documentation are distributed under the MIT License.
- MIT: FAES, Karpathy-inspired adapted guidance, Ponytail-inspired adapted guidance, the ECC integration bridge, and the vendored Cloudflare
security-auditskill (Cloudflare copyright preserved). - CC BY 4.0: Optional Unslop integration; upstream attribution is preserved in
skills/unslop/andintegrations/unslop/. - AGPL-3.0 (TruffleHog) and Apache-2.0 (Open Code Review): external tools only; no source is included in FAES.
Third-party component terms and upstream attribution requirements remain separate and must be preserved.
- Current version:
3.0.0 - Release history: CHANGELOG.md
- 19 core skills, 4 design-engineering skills, and 3 security/review skills (26 default)
- 2 optional skills: Unslop and release-hardening (BLA)
- 4 supported agent adapters
- PowerShell and POSIX installation and verification scripts