ara-research-manager

What problem does it solve? Research and coding sessions generate decisions, experiments, dead ends, and hypotheses that are quickly forgotten, leaving no auditable record of how a project actually evolved or which ideas came from the user versus the AI. ## Core Features & Use Cases - Post-Session Epilogue Recording: Scans conversation history only after a task completes, extracting decisions, experiments, dead ends, pivots, claims, and heuristics without contaminating the working context. - Provenance Tagging: Labels every entry as user, ai-suggested, ai-executed, or user-revised so reviewers can trace each assertion to its source. - Structured ARA Artifact: Maintains an exploration tree YAML, claims log, heuristics log, session records, evidence folders, and a staging area with automatic maturity promotion. - Use Case: After a long ML experimentation session where you tried FP16 training (failed), switched to GQA (worked), and formed a hypothesis about gradient instability, invoke this Skill to capture all of it into ara/ with correct provenance before closing the session. ## Quick Start Ask the AI to run the research manager epilogue to capture this session's decisions, experiments, and claims into the ara directory.