You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
kei has bounded diagnostic detail, but agents cannot control or continue most listings.
status, verify, and reconcile cap detail at 200 rows, and sync report JSON records omitted failure counts. However:
capped output has no cursor or supported continuation path
list albums and list libraries have no explicit bound or filter
manifest exports the complete catalog
agents must remember arbitrary --config paths because configs are not named or discoverable
#589 tracks a future local catalog query command. This issue should define shared bounds and configuration discovery that the query command can reuse rather than duplicating its query scope.
Proposed change
Add consistent controls for bounded reads:
--limit on list and detail-returning commands
deterministic ordering
a cursor or other explicit continuation mechanism where a complete traversal is useful
structured shown, truncated, continuation, and narrowing hints
filters only where they map to facts kei already owns
Build lightweight named configuration discovery on the existing TOML model:
list and show named configurations
select one with a stable root option such as --profile <name>
define precedence between explicit per-run arguments, environment, selected profile, and defaults
surface available names through agent-context
This does not need a second configuration format. A profile can remain a named pointer to a normal kei TOML file.
Acceptance criteria
List-style commands have a documented default bound and --limit.
Capped structured output tells the caller how to continue or narrow the next request.
Ordering and continuation are deterministic across repeated reads of unchanged state.
Explicit full exports remain possible for manifest.
Named configurations can be listed, inspected, and selected without remembering filesystem paths.
Existing --config behavior remains supported, with documented precedence.
Problem
kei has bounded diagnostic detail, but agents cannot control or continue most listings.
status,verify, andreconcilecap detail at 200 rows, and sync report JSON records omitted failure counts. However:list albumsandlist librarieshave no explicit bound or filtermanifestexports the complete catalog--configpaths because configs are not named or discoverable#589 tracks a future local catalog query command. This issue should define shared bounds and configuration discovery that the query command can reuse rather than duplicating its query scope.
Proposed change
Add consistent controls for bounded reads:
--limiton list and detail-returning commandsshown,truncated, continuation, and narrowing hintsBuild lightweight named configuration discovery on the existing TOML model:
--profile <name>agent-contextThis does not need a second configuration format. A profile can remain a named pointer to a normal kei TOML file.
Acceptance criteria
--limit.manifest.--configbehavior remains supported, with documented precedence.Relevant owners:
src/cli.rs,src/commands/,src/config.rs, state queries, and CLI/behavioral tests.Reference: 10 Principles for Agent-Native CLIs