What problem does it solve? Creating or modifying agent skills in this repository is error-prone because skills live in a dual-tree layout: the .agents tree is the authoring source of truth, byte-mirrored into the .claude tree by a sync tool. Editing the wrong tree, forgetting to register a skill in the sync manifest, or writing workflow-summarizing descriptions causes sync failures and degraded agent behavior. ## Core Features & Use Cases - Dual-Tree Authoring Procedure: Enforces authoring in the .agents tree, registering skills in SYNC_SKILLS, and verifying byte-identical mirrors with the sync tool's --check mode. - Skill Design Decisions: Provides a checklist for tier classification (core/specialist/optional), trigger-only description style, output schemas, self-test prompts, and overlap resolution with existing skills. - House Style & Guardrails: Defines frontmatter conventions, body structure (boundary intro, procedure, escalation rules), and a completion checklist. - Use Case: When adding a new skill for a recurring task, follow this workbench to decide its tier, write a trigger-only description, register it in the sync manifest, and confirm both trees match before committing. ## Quick Start Ask the agent to help author a new skill for the repository and walk through the tier classification, description style, and sync registration steps.
