AI Agent Workflow Collection

Scaffolds an Obsidian-compatible vault from the user manifest. Two modes — fresh-vault (creates Inbox/Logs/Plans/.coordination plus a personalized CLAUDE.md and System Backlog index) and retrofit-existing (walks an existing populated vault, renders a collision matrix, scaffolds only truly-new project candidates). Idempotent in both modes.

Scan the System Backlog for stale items, enforce lifecycle timeouts, auto-archive completed items, and produce a hygiene report. Use as a scheduled maintenance task or on demand. Trigger on: "backlog hygiene", "clean up backlog", "/backlog-hygiene", "stale backlog items", or any request to audit backlog freshness.

Deep research on a triaged System Backlog item. Reads the vault, the user's infrastructure, and external best practices, then writes an ideation brief plus a draft plan directory. Use when a backlog item needs research before planning. Trigger on: "research this backlog item", "backlog research", "/backlog-research", or any request to investigate feasibility of a system project idea.

Auto-classify a new or existing System Backlog item as NOVEL, DUPLICATE, OVERLAP, or DEFERRED. Use when a new idea is added to the backlog or when the user asks to triage an existing one. Trigger on: "triage this idea", "backlog triage", "/backlog-triage", or any request to classify a system project idea against the existing backlog.

Standing background processor that classifies and routes files dropped into the vault Inbox/ folder while the user works. One file → one classification → one routing decision → one audit log line. Transcript shapes route via the meeting note ingestor; project-shaped and ambiguous items remain in Inbox/ with "attempted" frontmatter so the user can triage manually.

Four-stage pipeline that takes a corpus of seed content (your existing notes), clusters by semantic similarity, proposes a vault taxonomy via LLM, renders a user-reviewable import plan, then runs a 3-step gate so you apply / edit / skip / abort before any vault writes happen.

Foundation-portable, source-agnostic transcript ingestor. Consumes a transcript file path (Otter VTT, Word, Zoom, generic LLM-export, or Granola JSON) and emits a structured meeting note (frontmatter + cleaned body) on stdout or to --output PATH. Zero vault coupling, zero People-registry coupling, zero engagement lookup.

Granola connector for meeting-note-ingestor. Wraps the portable ingestor with a friendlier flag (--granola-json) for users whose meeting capture pipeline is Granola.ai. The MCP fetch step is documented as an orchestration pattern, not embedded in this skill.

Verbal-first 5-section interview that captures identity, work context, vault layout, trust preferences, and privacy toggles. Produces a populated user-manifest.json plus an optional staged launchd plist in roughly 25 minutes. Hands off to /adopt when the vault is fresh.

Stage 3 of the content-seeding pipeline. Consumes a user-approved import plan and scaffolds project triads (PRD.md / Context.md / Updates.md) under each project candidate's proposed path, plus inbox-disposition entries for non-project candidates. Single batched preview gate; atomic on approve. Each generated file carries provenance frontmatter. Invoked by /adopt when an approved plan is detected post-onboarding.