01-core-provisioning-repos-constitution-and-modification-protocolisted

Trigger whenever you modify AGENTS.md, add or edit a skill, change provisioning-repos structure, register an MCP server, touch the sync scripts or gates, or do anything in the provisioning-repos — AND whenever you resolve ANY friction worth not rediscovering (a stale value, a missing path, an API/MCP capability gap, a gotcha, a wrong rule), because capturing that learning back is mandatory and autonomous. This is the CONSTITUTION: it defines the provisioning-repos architecture (the always-on vs on-demand split), the skill-authoring laws, how the gates physically enforce them, the propagation protocol, and the learn-from-friction loop.
markkennethbadilla/public-agent-provisioning · ★ 0 · AI & Automation · score 72

Install: claude install-skill markkennethbadilla/public-agent-provisioning

# Provisioning-Repos Constitution & Modification Protocol The four provisioning-repos (agent-provisioning-toolkit holds the brain: `AGENTS.md` + `skills/`) are the **single source of truth** for every CLI agent (Claude, OpenCode, Antigravity/agy). No agent home is ever edited directly. Its architecture is derived from Anthropic''s canonical skill-authoring guidance + the SKILL.md cross-agent open standard, NOT from how any older layout happened to be shaped. ## THE GOVERNING PRINCIPLE — "the context window is a public good" Two knowledge layers, and the entire architecture follows from keeping them separate: - **ALWAYS-ON** = `AGENTS.md` → synced to each agent''s instruction file (Claude `CLAUDE.md`, OpenCode `AGENTS.MD`, Antigravity `GEMINI.md`). Every token is paid on **every turn of every session, forever.** This layer is RUTHLESSLY minimal: ONLY the universal, every-task constitution lives here. - **ON-DEMAND** = `skills/`. Only each skill''s `name`+`description` preload (~30–100 tokens); the SKILL.md body loads when triggered; `reference/` files load only when needed. This is where everything that "matters only sometimes" lives. **The dividing law (gated):** a rule that applies to *nearly every task* → AGENTS.md. A rule/procedure that *matters only sometimes* → a skill, with a one-line `→ see <skill>` pointer in AGENTS.md so enforcement is preserved while the detail stays out of always-on context. Putting sometimes-content in AGENTS.md wastes context every