data-science-engineering-foundation

Resume by loading the session-state protocol, parsing the YAML state file, and validating the project slug, registry jobs, and lifecycle fields. The coach then announces the active job, paused bounded work, and active continuous context before asking whether to continue, switch jobs, or close.

Every registered job is exactly one of episodic, bounded, or continuous. Episodic jobs run Frame-Execute-Confirm per invocation, bounded jobs track phases and gates until a terminal recommendation, and continuous jobs persist as active background context anchored to a durable artifact.

State is stored at .copilot-tracking/ds/{project-slug}/session-state.md as a Markdown file containing one YAML block. The slug must match a kebab-case pattern, and the directory segment, project.slug field, and durable artifact identity must all agree before state is accepted.

The reconstruction protocol inventories durable customer artifacts as evidence, infers only supported facts such as project identity and artifact pointers, and presents a summary for explicit user confirmation. State is never recreated or replaced silently, and corrupt input is preserved when practical.

No. Every transition requires explicit user confirmation after naming the source job, destination job, carryover, and outgoing disposition. The source job is resolved by its lifecycle class using dispositions like completed, paused, or flushed, and a job_log entry records the change.

A write is blocked when the sensitive-content scanner reports any high-confidence finding, returns an error, or is unavailable. The gate passes only when the scan status is completed, data mode is true, and the high-severity count is zero; blocked writes name findings by category and location without exposing matched values.