Agent Teams lands. Nothing worth swapping.
First edition of the weekly sweep. Four buckets — PM, async comms, AI workflow, Obsidian. One verdict per item: Hold Trial Swap. The point is filtering out noise, not chasing every launch.
I asked the PM-Junior agent to run this scan every Sunday and only ping me if something genuinely beats what I'm already running. First go — not bad. We're in a good place.
Now back to work.
The headline
Anthropic shipped native multi-agent orchestration — Agent Teams — inside Claude Code at Code with Claude SF this month. One session acts as the team lead, coordinates teammates through a shared task list, each teammate running in its own context window. It is, almost line-for-line, the pattern I hand-rolled into the pm.plugin v2 walker a fortnight ago — itself a layer on top of the documented sub-agent pattern.
It's a trial, not a swap. The per-project CLAUDE.md anchors and the vault-walk discipline still belong to the existing walker — Agent Teams is the dispatch layer underneath, not a replacement for the orchestration logic on top. Worth a 30-minute poke on the next /pm run to see what collapses.
PM & dependency visualisation
Hold Businessmap, Teamhood, Planta. The 2026 multi-project Gantt round-up tools. None beat markdown + Mermaid + a vault-walker for a one-operator stack.
Hold Miro dependency mapping. Whiteboard angle. Excalidraw is already on the shortlist and is free and local.
Hold Datacore (Obsidian). "Dataview but faster." Only earns its place if I commit frontmatter discipline across every CLAUDE.md — which is a bigger decision than the plugin.
Async comms & status reporting
Hold StandIn. Built for the shift-handoff problem across timezones. I'm a solo operator. There is no handoff.
Hold Scrummer. Joins Meet/Teams calls, transcribes, structures standup notes. Granola and Gong MCPs already cover this for the meetings I actually run.
Hold Gitmore. Auto-generates standups from commits and PRs. My deliverables are markdown, decks, and reels — not a git-shipping cadence.
Hold Kollabe. AI-led async standups, free tier. No team to broadcast to.
AI workflow tooling
Trial Claude Code Agent Teams. Covered above. Native multi-agent dispatch inside Claude Code. Maps directly onto the pm.plugin pattern. Worth poking; not a swap because the per-project anchors stay canonical.
Trial Ruflo / Claude Flow v3.6. Open-source multi-agent orchestration platform for Claude Code. Thirty-one thousand GitHub stars and a stable release at the end of April. Read it alongside the Hermes Agent §3a entry before committing to either. If Hermes is the "always-on operational runtime" candidate, Ruflo might be the cheaper, better-supported alternative — Claude Code-native rather than a generic agent daemon. Don't install both.
Hold Dreaming & Outcomes (Claude Managed Agents). Self-improving memory and quality gating, hosted side. Wrong runtime for my CLI-first stack. Worth a note for any client engagement that ends up on Managed Agents.
Hold Claude Finance. Ten pre-built finance agents. Outside the portfolio.
Hold Oh-My-ClaudeCode. Team-oriented orchestration framework. I'm a one-man shop.
Hold Anthropic MCP connector (beta). Remote MCP without writing client code. Already covered by the existing MCP layer.
Obsidian & knowledge management
Trial Obsidian 1.10.0 — Bases formulas. Calculated fields, reusable formulas, multiple Bases views (table, cards, list, map). Hold for now — revisit if I commit to status: / blocker: / next: frontmatter across every CLAUDE.md. With that one move, Bases could render a live cross-project status board inside the vault, competitive with the deployed /status page.
Hold Make.md. Dataview-replacement plugin gaining traction. Same dependency on frontmatter discipline.
Hold Obsidian community site & plugin review automation. Ecosystem infrastructure, not a plugin to install.
Strong-swap recommendations
None this week. Boring weeks are normal. The point of this scan is to filter — three Trials and a headline is a healthy ratio for a first run.
What changes in the stack
Nothing this week. The toolbox.md and the hype-radar.md canon stay as they are. Next Sunday's scan re-runs and appends. If anything in the Trial column proves out — Agent Teams collapsing a chunk of pm.plugin, Ruflo replacing the Hermes question, Bases lighting up the vault — it lands in next week's Watch with a paragraph, not a line. The scan itself runs from the desktop launcher on a Sunday-morning trigger.
~/ClaudeWork/Claude/research-pm-tools.md