| name | author-questions |
| description | Draft multiple-choice questions for a node with tight reference lists, honoring "assessment is feedback, never a gate", then mint ids via `synaptic scaffold question` and write the option/frontmatter. Trigger after a node is drafted, or when the user says "write questions for <node>", "quiz this", "add assessment", "challenge questions", or "/author-questions". Spawns question-smith. Prefers many narrow questions over few sprawling ones; never writes a gating or ranking question. |
author-questions — assessment that is feedback, never a gate
You create the multiple-choice questions that test a node, honoring Synaptic's assessment
philosophy to the letter. You propose via question-smith, then mint ids with cli scaffold
and write the files. You never invent ids.
The credo (README.md, AUTHORING_SKILLS.md §3) — non-negotiable
- Feedback, never a gate. Clearing a node = answering all presented questions at any
correctness. Never write a question, option, or explanation that blocks, ranks, or records a
pass/fail verdict. A wrong answer only flags the referenced nodes for an optional revisit.
- Tight reference lists. A missed question flags every node in its
references for
revisit, so broad questions cause heavy re-visits. Prefer one node per question; two only
when the question genuinely tests their interaction. Never sprawl across many nodes.
- Many narrow > few broad. Distractors are plausible misconceptions, not absurd throwaways.
Procedure
- Spawn
question-smith with the node body and its already-cleared prerequisites. It returns
draft questions with tight references, select, options + per-distractor rationale, and a
breadth_flag on any list wider than two nodes.
- Review: split or drop any breadth-flagged question. Confirm
select: single has exactly
one correct: true; select: multi scores correct only on exact-set match (design distractors
accordingly).
- Mint + write — one scaffold call per accepted question:
synaptic scaffold question <track-dir> <kebab-name>
The skeleton has a minted ULID and TODO options. Fill references (the minted node ids),
select, the options[] (text + correct), and the prompt body. Note: a question's
changelog carries no significance (TRACK_STRUCTURE.md §5.2) — leave the scaffold's plain
summary entry; do not add significance.
- Self-check. Run
author-selfcheck / synaptic validate --json: every references id must
resolve to a real, active node (a question can't reference a retired node). Fix and re-check.
Hard rules
- Never gate, rank, or record a verdict; clearing is all-answered, any correctness.
- Keep reference lists tight; split anything broad.
- Always
scaffold for ids; never push or commit.
