Self-hosted control plane for governed AI agents: approval-first execution, persistent context, reusable profiles and skills, and auditable results.
This repository also configures its own agents. See what xeo-forge tells them →
Skill Claude CodeCodex
Verifies that a feature is wired through all four layers (schema, API route, runtime, UI) before it may be called complete. Use when adding or auditing any feature — mirrors the connectivity-map method that caught the v1.21 autonomy gap.
Skill Claude CodeCodex
Requires every new agent capability (subagents, sandbox tiers, future computer-use) to execute through the existing zone/permission-rule stack. Use when adding any capability that acts on the system — a capability with its own permission path is a design defect.
Skill Claude CodeCodex
Mandates immediate README disclosure of any discovered gap (unwired feature, unsupported limit, broken provider) before fixing it. Use the moment a gap is found during any work — the disclosure leads, the fix follows; silence or quiet removal is forbidden.
Skill Claude CodeCodex
Governs how Xeo Forge features are declared complete. Use when finishing any fix or feature — no claim of "done" is valid without a live probe (real API response, DB row content, or captured behavior) attached as evidence. Text descriptions of intent are never evidence.
Skill Claude CodeCodex
Constrains UI work to the shared design token vocabulary. Use when creating or modifying any component — new CSS values local to one screen are forbidden; every visual decision comes from the existing tokens (spacing, radii, colors, type scale) in globals.css.
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: