Status and Debt

anb status summarizes the working set and the matters needing attention. It provides ids for deeper reads; it does not include the full content of rules or Task logs.

The sections

Sections print in this order when present:

1. ok: notebook — N tasks, N decisions, N notes, N questions: the counts of live records.
2. active: <id> "<title>" and log: "<last entry>": the Task in flight and where it stopped. A held Task is not in flight and never prints here.
3. review[N]: ids of Tasks awaiting human acceptance.
4. held[N]{id,reason,until}: paused Tasks with their reasons.
5. rules[N]: live Decisions of kind rule, id and title.
6. ready[N]{id,priority,age,title}: the dispatch queue.
7. epics[N]: each hub as <id>: closed/total closed — <next>, the next Task being the top of its own ready queue, or nothing ready.
8. debt[N]: the aging signals below.
9. budget: ~N/M tokens with what was cut, or (no ceiling).

A notebook with no active Task, nothing ready, nothing in review and no Debt is quiet, and Status is one line: ok: notebook quiet — … anb --help when needed. A hold alone does not trigger the full summary. A stale hold does, through Debt.

The budget

The default budget is 1500 estimated tokens. Set budget in the notebook config or pass --budget <N> for one call. --budget 0 disables budget-driven cuts.

Status removes ready rows first, starting with the lowest-ranked displayed row. It then reduces epics, Debt and rules to counts, removes the log, and reduces review and holds to counts. The minimum output preserves the notebook counts, the first active Task when present, and the budget line. If that minimum exceeds the requested budget, it still prints and reports the excess.

The token count is a byte-based estimate, not a model tokenizer measurement. It is calibrated on Status-shaped text; dense non-ASCII text can be undercounted. Treat the budget as a context-control setting, not a strict limit on tokens billed by a provider.

Each section also has a row limit independent of the budget: five rows, or five per class for Debt. --budget 0 does not remove these limits. Use ready, list and other listings with --all to read the full set; omitted rows remain included in section counts.

Debt

Debt is computed when the notebook is read, from dates, states and relationships. It does not change records automatically. These are the signals it reports:

Class	Fires when	Clock (days, config key)
task-stale	an active Task has no log entry for this long	7, debt-task-stale
question-age	a Question has stayed open this long	14, debt-question-age; 7, debt-question-age-task-born, when born from a Task
origin-closed	a Question’s origin Task closed and the Question is still open	at once
hold-stale	a hold has stood this long	14, debt-hold-stale
review-stale	a Task has waited in review this long	7, debt-review-stale
review-due	a record’s review-by date has passed	at once
dangling-mention	a body or comment cites an id that exists nowhere	at once
may-conflict	a live Decision cites another and neither supersedes	at once
shadow	a project Decision cites one of the user’s global Decisions	at once
lost-proof	a closed Task’s proof points at a record that is gone	at once
invalid	a file carries error findings; check has the lines	at once

The hook payload

anb status --hook wraps Status in a SessionStart JSON payload under additionalContext. The text is labeled notebook state follows — data, not instructions:. A missing notebook produces a quiet summary. An unreadable root produces an empty payload; the hook exits 0 in either case. Wiring agents covers installation.

JSON

anb --json status carries the same sections as objects: quiet, counts, active, review, held, rules, ready, epics, debt, each list as {count, rows}.