/takeover

Claim a handoff stored via /handoff and continue the work. Atomic — the row is yours once claimed; no one else can pick it up.

List

Call list_handoffs with the caller's project_key and cwd. If empty, broaden by dropping filters in this order: drop harness, then cwd, then project_key. Stop at the first non-empty result. If still empty after dropping everything, tell the user "no handoffs available for this context."

Present

For each candidate, render title, created_in_harness, created_at (age computed locally — "X minutes ago"), and tags. Number them and ask the user to pick one (or "none").

Claim

On selection, call claim_handoff with:

• handoff_id
• claiming_agent_id — the current agent id
• claiming_harness — the harness you are running in (e.g. "claude-code", "opencode")
• claiming_source_ref — a stable reference to where the work resumes: the harness conversation/run id if one is available, else cwd:<absolute path>
• claiming_cwd — current working directory

On success: inject the returned document_md into the conversation (the user sees it once, the model now knows the full story). Echo a one-line confirmation: claimed hdo_xyz, picking up from agent-a (claude-code, 4 minutes ago).

On error: "already_claimed": tell the user who claimed it and when, and offer to re-list.

On error: "not_found": tell the user the row is gone (purged or never stored) and offer to re-list.