skill-composer

# skill-composer **P0 — Autonomous Multi-Step Workflow Chains** **Seal:** DITEMPA BUKAN DIBERI · 2026-03-27 --- ## Purpose The gap between B+ (81.8%) and A (90%+) is **autonomy** — the ability to chain a multi-step workflow without asking Arif for permission at every step. The `skill-composer` is the orchestration layer that: 1. Takes a sovereign goal (Arif's intent) and decomposes it into constitutional steps 2. Routes each step through the correct skill (research, plan, execute, verify, vault) 3. Enforces F1–F13 at each transition 4. Provides Arif with veto points at the correct moments (F13) 5. Logs everything and seals outcomes to vault **Why this is P0:** Without this, agents are slaves to the step-by-step prompt. Arif must babysit every research, every code change, every verification. The opencode catastrophe happened because an unanchored agent was given too much rope and no governance layer to stop it from hanging itself. --- ## Workflow Composition Types ### Type A — Research Chain ``` GOAL → deep-research → evidence-map → judgment → report ``` **Used for:** Understanding a system, investigating an issue, fact-finding **Constitutional gates:** F2 (Truth ≥0.99), F7 (Humility band), F3 (Tri-Witness) ### Type B — Build Chain ``` GOAL → architect-explainer → plan → constitutional-check → forge → postcheck-verifier → vault-writer ``` **Used for:** New features, configuration changes, deployments **Constitutional gates:** F1 (Reversibility), F11 (CommandAuth), F1...