What problem does it solve? Working in Claude Code involves harness-specific rules that are easy to get wrong: picking a model tier for a subagent, knowing that Claude Code cannot spawn another vendor's models, keeping .claude/skills pointer stubs in sync with canonical skill bodies, and reporting planning documents as clickable links instead of dead relative paths. This Skill consolidates those conventions so agents behave correctly before spawning subagents, wiring skills, or reporting documents. ## Core Features & Use Cases - Model Tier Resolution: Maps abstract tiers (deep, standard, mechanical, inherit) to the models the current session actually offers, independent of any written table. - Spawn Boundary and Agent Types: Clarifies that Claude Code spawns only Claude models, and distinguishes the Explore, Plan, and general-purpose agent types from model tiers. - Pointer-Stub Contract: Explains the .claude/skills/<kebab-name> stub requirement and regenerating stubs with sync_stubs.ps1 so skills trigger on current descriptions. - Clickable Document Links: Specifies absolute forward-slash hrefs for plans-repository documents, when relative links are valid, and when to use SendUserFile instead of the Artifact tool. - Use Case: Before spawning a subagent to execute an approved plan, resolve the plan's tier to a session model, choose the general-purpose agent type, and report the resulting plan as an absolute-path Markdown link. ## Quick Start Read this skill before spawning a subagent, adding a skill to a project repository, or reporting a plan document in Claude Code.
