AgentKit
DocsKitsCLI ReferenceDesktop App
BetaYou are reading docs for the beta channel (3.2.0-beta.10). Features may change before the next stable release.Switch to stable →

ak skills self-improve review

ak skills self-improve review list

List every screened skill with its latest decision, its recorded improvement candidates and their evaluation outcome, and the next commands.

Use ak skills self-improve review list to see every skill whose events crossed the review threshold, whatever the current consent. The command is read-only.

Usage

ak skills self-improve review list [flags]
ak skills self-improve review list
ak skills self-improve review list --json

The command takes no arguments and has no flags of its own. Shared flags are described in CLI conventions.

What it does

For each screened skill, the command shows its latest decision, the candidates ak skills improve recorded for it, each candidate's evaluation outcome (not_run until evaluated), and the next commands:

  • with no candidate, the /ak-dogfood review command;
  • for each candidate, ak skills evolution evaluate <id> while it is not evaluated, then ak skills self-improve review approve <id>;
  • ak skills self-improve review reject <skill> and ak skills self-improve review defer <skill>.

When the candidate store cannot be read, the command warns and lists the screened skills only. A candidate that fails verification is left out with a store.corrupt warning.

The command reads AGENTKIT_HOME/state/self-improve/events.jsonl, review.json, and the candidate records under AGENTKIT_HOME/state/skill-evolve/candidates/. It writes nothing and runs no model, script, or network call.

Output

Each text item shows the skill, its signal strength, its number of reasons and candidates, and its decision (undecided when none), then one line per candidate and one next: line per command.

JSON uses kind=skills.self_improve.review.list with these fields under data:

items[]: skill, digest, strength, reasons[], event_ids[], suggested_command,
         decision, nudged, evolution_candidates[], next_steps[]
decision: decision, digest, decided_at, until?, candidate_id?
evolution_candidates[]: candidate_id, identity, owner, proposal_only,
                        eval_outcome, apply_state
  • decision is null until you decide. Its decision is approved, proposed, rejected, or deferred.
  • nudged is true when the session-start line already showed the current finding and no decision silences it.
  • owner is user for a skill of your own personal kit and agentkit otherwise; proposal_only is true for agentkit.

Exit behavior

ExitMeaning
0Success, including an empty list.
1The AgentKit home could not be resolved (self_improve.home_unresolved), or the event store could not be read (self_improve.store_failed).
2Invalid flags or arguments (self_improve.invalid_input).
4Another writer held the event store lock (self_improve.store_busy).