Research Pipeline Runner

The runner turns one requested outcome into an auditable Workspace. Its leading word is continuation: every action must leave enough durable state for the next command or agent to continue without reconstructing the Run from chat.

Inputs

• The user Goal, including the target deliverable and constraints.
• An existing Workspace, or a path under workspaces/<name>/ for a new one.
• The selected Pipeline when already recorded in PIPELINE.lock.md.

Outputs

• A Workspace containing GOAL.md, PIPELINE.lock.md, UNITS.csv, STATUS.md, CHECKPOINTS.md, and DECISIONS.md.
• Pipeline-declared intermediate and target Artifacts.
• Run Evidence under .harness/ and output/.