Fygment is under active development. AI Skills governance is live for pilot workspaces; MCP setup, CLI workflows, lockfiles, and team automation remain preview or roadmap. See what is live →
v0.14 · controlled pilot

Govern AI Skills,
then grow the
agent toolchain.

Skills scattered across one-off harness folders. MCP servers in a JSON config nobody remembers editing. CLIs your team wrote and forgot about. Fygment is live today as a governed AI Skills registry and package workflow for controlled pilot teams. Local harness installs, MCP setup, CLI checks, lockfiles, and broader toolchain management are staged behind Preview and Roadmap labels.

$ npx fygment
Join pilot → Read roadmap →
● apache 2.0 Live now Skills registry Preview Codex + Claude Roadmap broader lifecycle
~/projects/payments-agent — fygment check
§ 01 · availability

Live now, Preview, and Roadmap stay separate.

The public promise starts with governed AI Skills. Setup automation and the larger control-plane vision stay labeled until the pilot evidence catches up.

Live now

Skills pilot
01Skills registry. Create, import, package, approve, export, and govern AI Skills in a workspace.
02GitHub and ZIP import. Bring in real Skill packages with validation, provenance, and checksum-backed storage.
03Workspace governance. Proposals, lifecycle states, audit history, ownership, and policies support the controlled pilot.

Preview

Setup path
01First writable targets. Codex and Claude-family setup lead install, check, update, and readiness hardening while broader harness support is staged.
02MCP setup. Metadata, guided CLI setup, and copyable configuration are visible as a preview workflow.
03Small teams. Invite flows, alerts, delivery settings, and pricing stay pilot-labeled while usage is proven.

Roadmap

Control plane
01Lockfile and sync. Skills-only lockfiles and local inventory come before arbitrary CLI management.
02Marketplace and delivery. Public marketplace, external delivery automation, and billing automation are not current pilot promises.
03Enterprise and security depth. Identity, compliance evidence, advanced deployment, and deep scanning stay long-term.
§ 02 · the problem

The agent toolchain has no toolchain.

Three or four harnesses on one laptop, each managing the same primitives in their own way. No shared inventory. No lifecycle. No idea what's installed where.

// what you have today

Ungoverned, scattered, manual.

drift
  • claude/projects/ 23 Skills, half stale, no version pin
  • ~/.cursor/mcp.json 7 servers, last edited 3 months ago
  • codex/config.toml re-wired by hand every time you switch
  • tools/agentctl that script Diego wrote — nobody owns it
  • source update important upstream change · nobody noticed
  • audit log does not exist
// fygment, instead

One inventory. One lifecycle. One policy surface.

Skills governance
  • ✓ Inventoried. Skills first; MCP setup and broader CLI inventory stay staged behind Preview and Roadmap labels.
  • ✓ Versioned. Current Skill packages are immutable; lockfile sync becomes the next reproducible layer.
  • ✓ Checked. Install state, source alerts, and policy status surface before they become surprises.
  • ✓ Policy-governed. Workspace approval, lifecycle, and resolver policy keep Skill rollout explicit.
  • ✓ Cross-harness. Fygment is built for many AI harnesses; Claude and Codex are current first writable Skill targets.
  • ✓ Traceable. Who installed what, when, and which workspace decision changed it.
§ 03 · CLI path

Skills commands first. Broader automation later.

The dashboard and CLI are paired during the pilot: the web app governs Skills, while local commands harden install, readiness, check, update, and the next lockfile workflow.

$fygment install <slug>

Preview for first writable Skill targets: install an approved Skill, ask before writing locally, and record install state.

resolvefan-out
$fygment list --paths

Preview inventory for managed Skills on supported first targets. Local paths appear only when you explicitly ask for them.

inventoryaudit
$fygment lock write

Roadmap path: write a Skills-only lockfile before claiming arbitrary MCP server or CLI package management.

lockfiledeterministic
$fygment sync --lock=fygment.lock

Roadmap path: reconcile approved Skills from a lockfile without removing unmanaged local files.

reconcileidempotent
$fygment check

Preview path: check managed Skills against the registry and fail closed on blocked, revoked, unsupported, or required-update states.

updatesgovernance
$fygment doctor

Preview path: report config, managed installs, target readiness, and local harness roots without exposing tokens.

diagnoserepair
§ 04 · harnesses

Skills first, broader harnesses staged.

A Skill is a Skill is a Skill. Fygment is built for many AI harnesses, with first writable local targets leading the same governed install model across the rest of the agent stack.

"If a Skill only works in one harness inside Fygment, we have failed." — design principle, Fygment manifest §1.2
Current and staged inventory
skill: code-review2.1.4
skill: writing-style0.9.1
mcp setup metadatapreview
mcp lifecycle managementroadmap
cli lockfile syncroadmap
arbitrary CLI packageslater
skills pilot
fygment
create · import approve · export install · check
First targets — Codex CLI and Claude-family harnesses · broader adapters staged next $ fygment readiness --target=codex
roadmap direction
fygment
Skills inventory live lockfile staged policy surface live
AI agent harnesses
Claude Desktopfirst
Claude Codefirst
Codex CLIfirst
Cursorroadmap
ChatGPT bridgeroadmap
Direction → governed setup spans AI harnesses · Skills live first · MCP setup path visible $ fygment readiness --target=codex
§ 05 · pilot path

One workspace, then a small team pilot.

The adoption path stays narrow on purpose: prove governed Skills with one pilot workspace, then expand into shared review, setup, and install posture for a small team.

01 stage one

Pilot workspace.

One engineer starts with a governed Skills registry, package export, and explicit approval history instead of private folders.

$npx fygment # guided setup
02 stage two

Invite a teammate.

Invite links and verified-domain prompts route teammates into owner/admin review without exposing private workspace details.

$fygment setup --workspace=acme
03 stage three

Small team checks.

Members, shared Skills, pending approvals, source alerts, and managed install risk become visible before the team expands.

$fygment setup --workspace=acme
04 stage four

Roadmap hardens.

Lockfiles, sync, marketplace, delivery automation, billing, and enterprise/security depth wait for proven pilot evidence.

$fygment check --target=codex
§ 06 · pricing

Free local basics. Paid AI imports and team control.

Start with a local skill inventory for free. Upgrade when imports, AI review, team governance, and enterprise controls become part of the workflow.

Solo

// individual workspace
Freeforever

Useful when you want a simple local setup before bringing in external sources or team workflows.

  • Local skill inventoryiTrack skills you create or install locally before setting up a team workspace.
  • Create and package skillsiTurn a prompt and supporting files into a reusable skill package.
  • Install/check in supported harnessesiCheck whether managed skills are installed and current in supported local agent tools.
  • Public templates and examplesiUse starter patterns and docs without connecting private source repositories.
Join pilot →

Advanced / Teams

// AI imports / teams
$14/seat/mo

For AI-assisted skill intake, private registries, and governed team rollout.

  • AI-assisted imports from GitHub, ZIP, npmiBring skills in from external sources and let Fygment inspect the package before rollout.
  • Imported skills version control and update checksiPin imported skill versions and check the main source for updates before changing what your team uses.
  • Security and compatibility reviewiReview manifest quality, target compatibility, risky files, and package structure before sharing.
  • Private team skill registryiKeep approved internal skills in one workspace-owned catalog instead of scattered folders.
  • RBAC and approvalsiUse owner, admin, and member roles to control who can submit, approve, and install skills.
  • Audit trail and install drift alertsiTrack changes, approvals, installs, and stale or unapproved versions across the workspace.
Talk to pilot support →

Enterprise

// enterprise/security
Custom

For regulated teams that need identity, deployment, compliance, and audit-depth guarantees.

  • Everything in Advanced / TeamsiEnterprise starts with the full paid team workflow, then adds identity, deployment, and compliance depth.
  • SSO (SAML, OIDC) and SCIMiConnect enterprise identity providers for sign-in and automated user provisioning.
  • Policy-as-code with PR reviewiManage rollout and approval policies through code review instead of ad hoc settings changes.
  • On-prem registry and air-gapped synciRun registry workflows in controlled environments and sync updates without open internet access.
  • SOC 2 Type II and ISO 27001 supportiSupport enterprise security reviews with controls, evidence, and roadmap alignment for compliance programs.
  • Unlimited audit retention and SIEM exportiRetain audit history for longer periods and export events into enterprise security tooling.
Talk to us →
§ end

Your AI Skills
deserve governance.

Join the Skills-first pilot, use the preview setup paths where they fit, and track the broader agent-toolchain vision on the roadmap.

Join pilot → Talk to engineering →
$ npx fygment