What problem does it solve? AI coding sessions lose context after compaction, mix fast and slow reasoning modes, and produce unstructured outputs that are hard to audit. This Skill imposes a layered operating protocol so every agent response is tagged, journaled, and verifiable. ## Core Features & Use Cases - Dual-Process Reasoning: Tags every response as #S1 (fast pattern-match) or #S2 (deliberate analysis) with explicit escalation rules when edits fail or uncertainty rises. - 3-Tier Memory Hierarchy: Journals decisions to a WAL file, persists design specs in docs/specs/, and treats code artifacts as ground truth, enabling context recovery after compaction. - CO-STAR Output Contract & Source Ladder: Structures non-trivial answers into Context/Objective/Steps/Thinking/Answer/References and ranks information sources from official docs (L1) down to model knowledge (L4). - Use Case: During a multi-file refactor, the agent reads the WAL and specs first, tags the session [CTX: refactor], escalates to #S2 reasoning, and emits a CO-STAR-structured plan with journaled decisions. ## Quick Start Activate the coprocessor skill and ask the agent to refactor the authentication module while following its session start ritual and CO-STAR output contract.
