/eklavya:quiz [topic]

Quiz the developer now. Follow the tutor skill for how to ask and grade — this command only decides what to ask about.

The developer asked for this, so always pass ignore_cooldown: true. The quiz cadence exists to stop Eklavya nagging; it must never refuse a request.

Topic given ($ARGUMENTS is non-empty): decide between a domain and concept slugs from the profile, not by guessing.

1. get_learner_profile with no domain. Its domains list names every domain there is; known is the list you must not ask about.
2. If the topic exactly names one of those domains, get_session_quiz_plan with domain: "<topic>". Otherwise pass every kebab-case word as a concept slug: /eklavya:quiz csrf jwt-structure is slugs: ["csrf", "jwt-structure"], and /eklavya:quiz caching is slugs: ["caching"]: a word that is no concept's slug is matched loosely by the planner, so pass it as given. Either way pass ignore_cooldown: true. The plan already drops what they have mastered, orders prerequisites first, sets each tier_to_ask, and hands you asked_before so nothing repeats.
3. If that plan is empty and the topic was not a domain, nothing by that name is due: the concept is unknown, or already mastered and not due. Say that in one line, offer the closest domain from step 1's list, and name /eklavya:learn <topic> for a lesson that adds new concepts. A quiz never mints concepts: do not call upsert_concepts, do not offer to build concepts from this repository, and do not invent questions.

No topic: quiz this session's work.

1. get_learner_profile.
2. get_config, then get_session_quiz_plan with ignore_cooldown: true and max set to the config's max_questions_per_task. Both arguments are there for the same reason: on the default as-you-go cadence a plan is capped at one question, because Eklavya interrupting is not allowed to become a pile-up — but the developer typing /eklavya:quiz asked for a round, and a request is never refused or trimmed. A named topic (above) is already treated as a request; this branch has to say so.

3. If questions_needed is 0, say why in one line and stop. The reason tells you which:

   • already_covered → "Everything from this session has already been asked about. /eklavya:quiz <topic> to go wider."
   • nothing_logged → "Nothing logged this session yet, so there's nothing grounded to ask about."
   • no_candidates (topic mode) → for a domain, it is fully mastered and nothing is due; name the closest domain that is not. For slugs, follow step 3 of the topic branch.
   • quiz_disabled → "Questions are off for this project. Memory is still recording — /eklavya:mode turns the questions back on." Do not say "Eklavya is off": it is not, and saying so is what sent people hunting for a bug that was a setting.
   • session_off → they silenced the questions for this session, and then asked for a quiz. Say so and offer the one step back: "Questions are off for this session — say the word and I'll turn them back on." Turn them back on with set_config, scope: "session", quiz: { enabled: true } if they agree, then run the quiz.
   • panel_question_open → a question is already waiting in the side panel: "There's already a question waiting in the panel — answer it there." Ask nothing else.
   • no_topic → focus is learn with nothing set; ask what they want to learn, then /eklavya:mode learn <topic>.
   • topic_unknown → the graph has nothing matching their topic; offer the closest domain rather than inventing questions.

   If the plan carries pending_elsewhere, add one line after the reason naming those projects by folder name with their counts, most first: "Pending in other projects: eklavya (12), PiDom (3) — run /eklavya:quiz there." Say nothing about other projects when the field is absent.

A named topic overrides the configured focus for this quiz only. Someone who types /eklavya:quiz caching wants caching now; it is not a request to change their standing setting. Do not call set_config — /eklavya:mode is for that.

Then ask one question at a time at each concept's tier_to_ask, following the plan's framing field, never repeating anything in asked_before, grade every answer with record_attempt and tell them whether it was right before the next question, and close with one line: what moved, and what comes back for review when.

When the plan says presentation: "panel", show the first question with present_question as the tutor skill's references/panel.md describes and stop there: the panel records each answer and shows the verdict, so do not record_attempt or judge it yourself. The panel's Next button asks for the next question of this round; present the next item when it does, and close with the same one line once the round is over.