ralphlisted

# Ralph — Graph-Orchestrated Backlog Loop **Usage**: `/bs:ralph [--target-dir <path>] [--inline] [--until "N items"] [--scope all|feature|bug|effort:S|effort:M] [--section all|high|medium|low] [--quality auto|95|98] [--classic] [--wt] [--parallel] [--reflect-depth standard|deep] [--speculate auto|always|never] [--score-threshold 0.7] [--evidence-dir .claude/ralph-next] [--max-retries N] [--max-ci-retries M] [--max-quality-minutes N] [--no-compact] [--dry-run]` **Defaults**: 10 items, all scopes, all sections, auto quality, standard reflect depth, speculate auto, score threshold 0.7 **Backlog sources (Linear OR inline OR both):** 1. **Linear** — `mcp__linear__list_issues(filter: { state: { name: { eq: "Backlog" } } }, orderBy: "priority", first: 1)`; mark done with `mcp__linear__update_issue(id, stateId)`. 2. **Inline** — markdown bullets / numbered items in `$ARGUMENTS` are auto-detected and synthesized into ephemeral backlog items (not written to Linear). 3. **Both** — when both are present, the inline list is appended to the Linear queue **for this run only** and gets the same state-machine treatment. `--inline` forces inline-list interpretation if auto-detection misfires. State machine: `PICK -> IMPLEMENT -> QUALITY -> REFLECT -> DECIDE` **Arguments received:** $ARGUMENTS ## Fresh-campaign and budget contract Do not fork this long-running loop from the caller's transcript. The initial invocation runs in its current session; unattended runners start one fresh, non-