The working layer¶

The vaults hold what you know. The working layer holds what you are doing about it. It follows the same rule as the vaults: unified reading, segregated writing.

Three trackers, one picture¶

Tracker	Answers	Written by
beads (`bd`)	What is the agent doing? What is ready, what is blocked?	AI agents
the vault	What are we trying to close, and why?	you + agents
Linear (optional)	What do stakeholders see?	you

beads is a local, git-backed task graph (Steve Yegge's bd). No account, no cloud. Agents queue, claim, and close work here; the ready queue is derived from explicit dependencies. Tasks carry a domain: label so work filters by the same five domains as the vault.
the vault is where planning lives as prose: _Index/Open Loops.md, Architecture/decisions/, and Weekly/. It is the narrative both the agent and you read at the start of a session.
Linear is an optional adapter for a human tracker, kept separate so agent execution and stakeholder tickets do not bleed into each other. It is on the roadmap, not required.

The loop¶

./standards/skills/plan/plan.sh   # ready beads + the vault's open loops, side by side

Pick a ready bead whose domain matches an open loop.
Claim it, announce the claim, do the work.
Close it with a one-line result, and update the open loop.

See the graph¶

The work-graph dashboard renders the same task graph in a browser: issues by project, dependency edges, and the ready queue. It reads beads through bd export only.

Why keep them separate¶

Agents and humans work on different horizons and need different views. An agent needs a dependency graph and a ready queue; a stakeholder needs a status and a milestone. Mixing them makes both worse. Segregated writing keeps each tracker honest; unified reading (everything is retrievable, and plan shows both at once) keeps them connected. The conventions live in standards/canonical/work-tracking.md.