loobster

Reference only (do NOT invoke as an action): token-reduction conventions (subagent isolation, artifact compaction). Applied throughout the other skills.

Reference only (do NOT invoke as an action): the RICE scoring convention for the goal-loop backlog. Read by the loop/make-plan skills.

Implement a single planned issue, optionally inside a bounded autonomous loop. Use to execute one unit of an approved plan.

Run a continuous goal-loop: work a prioritized RICE-scored backlog toward a standing goal, cycle after cycle, consuming/emitting signals. Use for ongoing autonomous improvement toward a goal; not for a single discrete task (use run).

Run the enabled compliance checklists (any of HIPAA / HITRUST / ISO 27001 / SOC 2) against the diff, separating code-verifiable findings from organizational controls. Always run before committing a sensitive change.

Emit or consume cross-loop signals (observations / frictions / opportunities) in the shared signals/ store. Use to record a finding for other loops or teammates, or to read relevant ones before deciding. PHI must never go in a signal.

Verify UI changes by capturing Playwright screenshots and attaching them to the PR (GitHub-native, no third-party hosts). Use when a change touches the frontend.

Resume a paused, interrupted, or crashed RePPITS workflow by reconstructing state from tasks. Use when continuing prior workflow work after a break.

Break the chosen proposal into ordered issues/tasks with dependencies. Use after a proposal is chosen, before implementing.

Generate up to two solution proposals grounded in prior research (with architecture diagrams). Use after research, before planning.

Document the existing codebase exactly as it is today (no suggestions, no fixes). Use as the first phase, before proposing changes.

Review all uncommitted changes (the Test phase) in a SEPARATE verifier — never self-review. Use before securing/committing.

Run the full RePPITS workflow (Research -> Propose -> Plan -> Implement -> Test -> Secure) with risk-tiered approval gates. Use to build or change a feature end-to-end. Not for tiny one-off edits.