Status and Debt

anb status summarizes work: active Tasks, reviews, holds, ready work and open Questions. anb recall opens a session with that summary, the session’s focus and the standing rules from every notebook it reads. Both provide ids and commands for deeper reads. Neither chooses a session’s Task merely because it appears first.

The sections

Status carries the following fields in both TOON and JSON:

Field	Meaning
quiet	no active work, review, ready work, untaken pool, open Question or Debt
by	the identity filter, when narrowed
team	a command to see the whole team’s work, when narrowed
more	the same Status without an output ceiling
counts	live Task, Decision, Note and Question counts for the whole notebook
active	active, unheld Tasks, each with its latest log entry when available
review	Tasks waiting for review
held	paused Tasks with reason and optional resumption date
ready	startable Tasks in queue order
untaken	the ready, unassigned pool count and its read command
questions	open Questions, own first and then oldest first
debt	maintenance signal count and its read command
budget	estimated output size and the requested ceiling

Each work section is {count, omitted, rows}. A quiet notebook still returns a structured summary. A hold alone does not make it busy; a stale hold does, through Debt. Decisions and Notes do not appear as work rows.

Whose work

Status is the whole team’s by default, with your work first. --mine uses ANB_BY or git user.name; --by <name> selects a colleague. Work follows the holder or the person it is addressed to, not merely the author who created it. An unassigned Task remains in the pool.

A notebook configured with scope: mine narrows Status without a flag. --team widens one call. The reply exposes the resolved identity and a widening command so an empty personal result cannot be mistaken for an empty notebook. Without an identity, personal work filters return an actionable refusal; the native hook carries read failures as diagnostic context.

Narrowing changes visibility, not validity. A Task blocked by a colleague’s work stays blocked when that work is outside the selected scope.

The budget

Status and Recall default to 1500 estimated tokens. Set budget in notebook config; Status also accepts --budget <N> for one call. Status --budget 0 disables cuts and includes every row. Recall --all lifts its row and text bounds.

The CLI selects one reply document for both formats and measures its actual TOON encoding, including the trailing newline. budget.limit is the requested ceiling, or null when unbounded; budget.spent is the resulting estimate. JSON carries those same values and selections.

A bounded Status starts with at most five rows per section. To fit the ceiling, it removes ready rows first, then Questions, holds and reviews, then active log text and extra active rows. Counts and omission markers remain. The minimum retains notebook counts, section counts, navigation and the first active row when present. That row is an ordered summary, not remembered session focus. If the minimum exceeds a tiny requested ceiling, spent reports the excess honestly.

Recall budgets its entire composite once. A session opening carries the work, the focus as an id and a title, and every live rule and drift Decision from the project, personal and global notebooks. Rules apply to any work the session takes on. A drift is an agreed exception to a rule, and it comes along because the rule alone would mislead. Other Decisions and Notes are read when the work cites them. Each source counts them as other and gives the more command that lists them. Selection takes turns across the sources, so a large project cannot take every row. This order does not decide which instruction wins a conflict.

To fit the budget, an opening shortens memory bodies to 256 characters first. If that is not enough, it drops the bodies and prints the memories as a table of scope, id, title, kind and read command, with bodies-omitted set. Work rows are reduced after that. Memory rows go last: first from sources that still show more than one, then from any source. The title of a rule usually states the rule, and the read command gives the body.

recall "<phrase>" searches instead. It returns every live Note and Decision that matches the phrase, from every source, without work or focus. The reply lists every match even when that exceeds the budget. To fit, it shortens the bodies, then drops them, and cuts invalid-file rows last.

The estimate is byte-based, not a provider’s tokenizer. It controls context size but does not predict billed tokens exactly. Core callers receive complete ordered models; output selection belongs to the CLI.

Debt

Debt is computed from dates, states and relationships when read. It never changes records automatically. anb debt lists signals in the following order, bounded to 20 rows by default and lifted with --all.

Code	Condition	Default clock or trigger	Row fields
task-stale	an active Task has no recent log entry	7 days; debt-task-stale	id, days
question-age	a Question remains open	14 days, or 7 when born from a Task; debt-question-age, debt-question-age-task-born	id, days
origin-closed	an open Question’s origin Task closed	immediately	id, origin
hold-stale	a hold has stood too long	14 days; debt-hold-stale	id, days
review-stale	a Task remains in review	7 days; debt-review-stale	id, days
review-due	a record’s review date passed	immediately	id, date
dangling-mention	body text cites an id absent from the same notebook	immediately	id, target
lost-proof	the host establishes that a linked commit or report file is missing	immediately	id, proof
invalid	a record has error findings	immediately	file, errors

Shared tags and citations do not prove that two Decisions conflict. Personal notebooks do not validate shared references or change shared Debt. Resolve differences explicitly in the relevant records.

External evidence checks inspect filesystem paths and git commits without modifying them. Pull request URLs are not checked. An unavailable git query cannot establish that a commit is missing; absence of lost-proof is not proof that work was verified.

The hook payload

The installed anb hook adapter wraps a budgeted Recall document in the host’s SessionStart JSON envelope, under additionalContext. It identifies that content as notebook data, not instructions, and carries the current session’s focus when available: its id, title and read command. Unreadable state produces a diagnostic rather than an empty-memory report, without failing the session hook. Wiring agents covers installation, session input and environment propagation. status is a work query; it has no hook flag.

JSON

anb --json status has the same document described above. A row carries by, taken-by and to when present. Debt rows carry code, the fields in the table and a human-readable line; programs should use the named fields.

Recall adds work, focus, memories, sources, count, omitted, invalid, more and the single outer budget; a search leaves out work and focus. focus carries id, title and the read command. Each memory names its scope and read command. Each source carries {scope, count, omitted}, and on an opening other and more when it holds knowledge the opening left out; omitted body characters are reported on the individual body. Equal ids in different notebooks remain distinct; personal guidance is not merged into team knowledge.