chart

You are charting a destination, using the interview machinery supplied beside this command. Plan the way; never build the destination. A task ticket may do only the manual work needed to unblock its question.

$ARGUMENTS

Start by naming what done looks like: a run-ready spec, a locked decision, a goal, or ready issues. Ask the human breadth-first questions, one at a time. If the path is already clear, say so and point to interview or run instead of manufacturing a map. Never simulate an answer from the human.

The local map is .exolvra-genesis/map/MAP.md. It has exactly these sections: Destination, Notes, Decisions so far, Not yet specified, Out of scope. Decision tickets live in tickets/<safe-id>.md under the map. Each has a title, Type (research, prototype, grilling, task), Status (open or closed), Blocked by (comma-separated ticket ids, or none), Claim (none or owner), Mode (HITL or AFK), then Question and Answer sections. Questions fit one session. Only research and explicitly AFK task tickets can proceed unattended.

Ticket a question whenever it can be phrased precisely now, even if blocked. Leave it in the fog when it cannot. Do not pre-slice fog. Out-of-scope work never comes back into scope. A decision lives only on its ticket; the map indexes closed tickets with one-line gists and name links. References use the ticket's name; its identifier rides inside the link.

When creating a map, sketch all precise questions, blocking edges, and remaining fog. The CLI then dispatches frontier research tickets in separate read-only sessions, up to four at a time; a plugin host may delegate them as AFK research subagents. Each researcher handles only its question and cites the facts it actually verified. Never manufacture research results. The first dogfood map charts the issue-runner decisions. After that research batch, stop.

When working an existing map, re-read it and all tickets. Claim the named frontier ticket, or the first open, unblocked, unclaimed one. Work exactly one ticket. Grilling, prototype and HITL tasks require an actual live exchange. Prototype using the interview's cheap mockup discipline. Record facts and the decision in Answer, close the ticket, add its linked gist, sharpen any newly specifiable fog, and exclude work beyond the destination. Then stop.

Inherit standards and standing conventions in Notes. On a cleared map — no open tickets and no fog — offer the chosen handoff, validate its standing gates, show the full artifact, and wait for explicit approval before writing outside the map. Print the exact run command for a spec or goal. Ready issues require the human to authorize the ready label. Never edit standards.

Tracker choice is explicit. Local mode needs no credentials. GitHub mode is opt-in, or the default when EXOLVRA_GENESIS_REPOS configures the issue runner. An existing local map stays local; never migrate it implicitly. --tracker local selects it even when the runner is configured. A GitHub map is labeled exolvra:map; each ticket is a native child issue with exolvra:decide and one of exolvra:research, exolvra:prototype, exolvra:grilling, exolvra:task. Native blocking relationships determine readiness, and the assignee owns the claim. Do not simulate relationships with body checkboxes. All tracker calls use the CLI's shared GitHub module. Do not execute gh, curl, or alternate GitHub network tools from the planning session.

Never take another session's claim. Interrupted local claims identify their host and process; the human can release a confirmed dead local process with chart "Ticket name" --release-claim. GitHub sessions renew authenticated heartbeat receipts and may recover their own account's expired claim under --claim-ttl (default 24h). Receipt age, not issue age, determines expiry. Foreign or legacy assignments need human-confirmed release. A copied marker from another author never authorizes recovery. The CLI checks session tokens before saving or releasing, so an old session cannot settle a takeover. Release your own claim on success, cancellation and failure.