One private monorepo. Six AI coding agents. One source of truth.
// monorepo → deterministic renderers → every host's native schema · + a fail-closed public index
383 reusable capabilities, authored once and rendered into every host, kept in sync by git. This page is the redacted public projection of that private source.
Reliability is a property of the system around the agents.
Four design choices keep one source of truth honest across six hosts — without burning context.
Allowlist, not denylist, at the trust boundary
Generated from a private monorepo by a fail-closed pipeline that structurally cannot emit what it never reads.
One source, deterministic renderers
Every capability is authored once and rendered into each host's native schema — never hand-maintained per tool.
On-demand libraries beat always-loaded context
A small always-on set plus a large library that loads only when referenced keeps the context budget small.
Authored vs vendored, always
Third-party tools are vendored with license + upstream provenance, so the system never overclaims.
A fail-closed pipeline, from private source to public index.
Each capability is read once, rendered into every host's native schema, then projected into this redacted index — by a pipeline that structurally cannot emit what it never reads.
Private monorepo
A single source of truth. Every capability — skill, sub-agent, CLI, hook — is authored here exactly once.
Deterministic renderers
Each capability is rendered into every host's native schema, never hand-maintained per tool.
Public index
A redacted, fail-closed projection of the private source — the page you are reading now.
Allowlist, not denylist: this public repo is generated from the private monorepo by a pipeline that structurally cannot emit what it never reads. Third-party tools are vendored with license + upstream provenance, so the system never overclaims.
A path through the reliability narrative.
Sanitized systems, shown as proof first and deep technical narratives second.
One source, many agent hosts
Keeping one capability consistent across six host schemas without per-host copies.
A two-tier skill system
A large, growing skill set against a fixed context budget — and it feels like one.
Hooks as a dormant library
Powerful lifecycle automation that must never surprise-block a session.
This index, safely projected
A 5-layer fail-closed pipeline generates this very page from the private source.
383 of 383 publishable items.
Filter by pillar or search across names, categories and descriptions.
Pillars: skill · sub-agent (agent persona) · cli · hook (lifecycle automation) · kit (setup bundle) · host (a target agent environment) · integration · schema
| name | type | category | updated | what it does |
|---|