/study-next — open the next session

The plan is a queue, not a calendar. You open the first session that is not closed, whatever today's date is. A missed day means the queue did not move: there is nothing to "catch up" except the dates, which you re-project.

$ARGUMENTS may be a plan slug, to open a plan that is not the active one.

Step 0

SHARED="$HOME/.claude/skills/study-shared"; SS="$SHARED/scripts/study_state.py"
PLAN=$(python3 $SS registry resolve $ARGUMENTS) || exit   # exit 2: ambiguous → ask which; 3: none → suggest /study-new
TODAY=$(python3 $SS today)
python3 $SS next "$PLAN"

Read $SHARED/references/frontmatter.md if you have not this session. Read $SHARED/references/recovery.md only when behind_sessions > 0 or ahead_sessions > 0. Take dates from $TODAY, never from date.

Step 1 — drift first

From the JSON: behind_sessions, ahead_sessions, buffer_left, horizon.

• behind > 0 → apply recovery.md for the plan's horizon. Flexible/open: re-project and say it in one line, quoting the new planned end. Fixed: propose buffer → double → trim, show the resulting calendar, apply only on approval. After any re-projection run python3 $SS check "$PLAN" and handle milestones as recovery.md says.
• ahead > 0 → say so; offer to pull forward, go deeper, or finish early. Change nothing unless asked.
• The next session is a buffer and there is no drift → offer to skip it (set … status=closed) or use it for review. The buffer is a reserve; it is not consumed by default.

Step 2 — a long pause gets a warm-up

If days_since_last_activity in the Step 0 JSON is 7 or more, offer a 5-10 minute warm-up before the new session. Use that field, read before marking anything: it is the latest open, evaluation or close date across all sessions, whereas a single actual_date only says when a session was opened (it can stay open for days through a failed evaluation and its re-evaluation). The warm-up covers the previous session's 2-3 core ideas from its notes region, or two quick questions on them. Offer it; do not impose it.

Step 3 — act on the session's status

• pending:
  1. If the session body still has unfilled template content ({{…}}, check reports placeholder, or sections with no real content), do not mark it studied: say it is incomplete and offer to complete it with study-new's rules (verified links, backed claims, a specific analogy).
  2. Otherwise mark it first, then compose the reply, so the state you quote is already true:

     python3 $SS set "<file>" status=studied actual_date=$TODAY && python3 $SS refresh "$PLAN"

  3. Present the session: topic, why it matters, checklist, the analogy verbatim (it was written for this exact confusion), readings with times, videos with the same weight as the readings (their duration shown, what they cover and what they do not), practice for its level. Do not re-verify links: they were verified when the plan was created.
  4. Close with "When you are ready, /study-eval. Mark guesses with ?." (in the plan's language).
• studied, eval_passed: null → "You opened this on <actual_date>. Ready to evaluate?" Offer a quick recap of the checklist if they want it. Do not deliver the whole session again.
• studied, eval_passed: false → review the items in the session's weak region one at a time, then hand over to /study-eval for the re-evaluation, recommending it at the start of the next sitting (protocol.md). If the weak region still holds the template's hint text (plans made before iteration 2), rebuild it first from the result region and PLAN.md's concepts.
• evaluated → "This one is evaluated and passed. /study-close to write your notes and move on."
• eval_type: none and studied → there is no evaluation; offer /study-close directly.

Practice that fails

If the user reports that a live command failed for permissions, treat it as data: note it in the session's weak region, downgrade that area's practice_level in the remaining sessions (set … practice_level=sandbox or none), and stop proposing that access ($SHARED/references/practice-levels.md).

Language: files in the plan's language, conversation in the user's (language.md).