sussdorff

Orchestrate parallel implementation of multiple beads across cmux panes in dependency-aware waves. Use when implementing a whole feature area, dispatching multiple beads at once, or running parallel cld -b sessions. MUST USE when user says "wave", "parallel beads", "implement all beads for X", "start the beads", or references implementing more than 2 beads at once. Also triggers on "cmux dispatch", "multi-bead", "wave orchestrator".

Audit a repo for ADR hoist debt and pre-trinity packages. Triggers on: adr gap, hoist debt, adr audit, pre-trinity packages, architecture drift.

Troubleshoot beads Dolt failures: `bd dolt` push/pull errors, merge conflicts, embedded/shared-server mode problems, remote auth issues, local DB recovery, re-clone, reflog restore, broken `.beads` Dolt config, and brew-services Dolt lifecycle. Use whenever beads sync or Dolt infrastructure is failing — symptoms like "beads do not sync", "cannot push issues", "no common ancestor", "not supported in embedded mode", "Dolt server unreachable", or "lost local database". Read the beads changelog first. Do not use for normal beads tracking (`bd create/ready/close`), standalone Dolt databases, or regular git push/pull.

Create and review agents (subagents). Use when creating specialized AI assistants, reviewing/auditing existing agents, deciding between agent vs skill vs command, or asking about agent best practices, multi-agent pipelines, and model selection. MUST BE USED when the user says "create agent", "new subagent", "review agent", "audit agent", "agent vs skill", or asks about multi-agent architecture.

Query per-bead token cost metrics from the local agent metrics database. Shows cost table, impl:review ratios, most expensive beads, cost trend over time. Triggers on: bead metrics, token cost report, bead costs, /bead-metrics.

Structured 4-stage retrospective for completed beads: Requirements vs Strategy, Strategy vs Execution, Iteration Review, Workflow Audit. Creates Trap/Pattern/Decision artifacts. Use when debriefing a bead, capturing session learnings, or running a compound retrospective.

Smart bead creation with type guidance, size routing, and auto-scenario generation for features. Use when creating beads, adding work items, or when user describes something to build/fix/improve. Triggers on bead erstellen, create bead, neues bead, new bead, bd create.

Guided planning dialog for larger initiatives producing Beads epics with sub-tasks. Use when planning features or multi-task initiatives needing structured breakdown. Triggers on epic init, plan epic, create epic, plan feature, break down feature.

use when: saving a single URL (article, YouTube video, blog post, documentation page) to open-brain memory as curated_content with provenance metadata. NOT for: bulk import of multiple URLs (run once per URL); local files (use `summarize ./file` + `ob save` manually); JSONL or Markdown batch import (use ob-migrate). boundary: ingest-content is one-URL-at-a-time with content-type detection; ob-migrate is for batch/interactive memory writes from files or prior context.

use when: bootstrapping or extending open-brain memory by importing facts from prior conversation context, a JSONL export, or Obsidian/Markdown files. NOT for: ingesting URLs (use ingest-content), or saving a single observation (use `ob save` directly). boundary: ob-migrate is for batch + interactive memory writes; ingest-content is for one URL at a time; ob-search is read-only.

use when: manually reviewing open-brain memories with human-in-the-loop approval — classify keep/merge/archive/delete/promote, route promotes to marketplace primitives via the matching *-forge skill, scaffold beads from observations. NOT for: scheduled or silent lifecycle maintenance (use memory-heartbeat); writing new memories (use ob-migrate or ingest-content). boundary: ob-triage is interactive (AskUserQuestion per action); memory- heartbeat is autonomous (cron-triggered, silent when QUIET).

use when: scheduled/cron-driven open-brain maintenance — detects time window (WORK-HOURS / END-OF-DAY / WEEKLY / QUIET) and runs the matching silent action (lifecycle pipeline, daily digest, weekly summary, provenance check). NOT for: manual memory review with approvals (use ob-triage); writing new memories (use ob-migrate or ingest-content). boundary: memory-heartbeat is autonomous and silent in QUIET mode; ob-triage is interactive with AskUserQuestion per action.

Search open-brain memory for past observations, session summaries, and learnings. Use when: "search memory", "what did I learn", "past observations", "past sessions", "what happened before", "previous context", "recall", "remember", "erinnere dich", "was habe ich gelernt", "letzte Session", "memory context".