syncing-agent-skillslisted

# Syncing Agent Skills ## Overview This skill maintains the repo's dual-harness skill strategy: - **Shared skills** live canonically in `.agents/skills/` - **The projection engine** (`@dorkos/harness`, driven by `dorkos harness sync`) projects the canonical layer to every enabled harness — Claude Code, Codex, Cursor, Gemini, Copilot - **`.agents/harness.manifest.json`** declares the exceptions: Claude-only skills, tool-specific wrappers, command mappings, and instruction/hook projections - **Claude Code** consumes shared skills through per-skill symlinks in `.claude/skills/`; **Codex** discovers them directly from `.agents/skills/` The goal is to keep shared expertise in one place and let the engine, not by-hand symlinking, keep every harness's view aligned. ## Primary Workflow: the Sync Engine The engine is the source of truth for projection mechanics. Run it from the repo root: ```bash dorkos harness sync # report drift, touch nothing (same as --check) dorkos harness sync --fix # realize the plan on disk (symlinks, scaffolds, generated files) dorkos harness sync --check --harness codex # narrow to one harness (claude-code|codex|cursor|gemini|copilot) ``` - `--check` (the default) prints the per-harness projection plan and exits `1` on drift — safe to run any time, ideal after creating/renaming/moving a skill - `--fix` applies the plan: creates missing symlinks, scaffolds wrappers, regenerates projected files, and reports anythin