education: add quiz-me skill — post-work comprehension verification (HITL, non-gating by default)

[kyle-sexton]

Purpose

Third sibling in the education plugin (teach = multi-session coach, explain = one-shot altitude drop): after Claude completes a change, generate an HTML report of what was done (context, intuition, decisions) with a quiz the user answers — verifying the HUMAN absorbed the work. Source practice: Thariq Shihipar (Anthropic), "Field Guide to Fable" talk + Peter Yang episode.

Value proposition (nail the metadata to this, not to artifact quality)

Verbatim reasoning from the talk (~13:57): have Fable "quiz you about what happened, to confirm you understand the work and can represent it when creating or merging a PR" — key to "staying in the loop." Episode failure mode it counters: "people still glaze over the plans and explainers."

1. Representation accountability — the PR author must be able to defend the change to a reviewer.
2. Loop retention — the user's mental model stays current, keeping their future prompting good.
3. Late intent-mismatch detection (secondary) — a failed quiz can surface "that's not what I intended" pre-merge.

Design decisions (locked in interview session claude-loop-practices, 2026-07-21)

• Object under test is the human, not the artifact. verification:confirm owns artifact verification; this skill never overlaps it. Composition by pointer only (e.g. optional stage in session-flow workflow gate lists).
• Non-gating by default. Comprehension aid, not a required gate. Opt-in gating via userConfig, e.g. quiz_policy: off | on-request | always | above-threshold (threshold keyed to change size / blast radius).
• HTML report + embedded quiz per the canonical prompt pattern; markdown fallback per project convention.
• Name via the codified naming grammar (quiz-me is the working label, not locked).

Sources (tier-1 + digested)

• Written canon with verbatim prompt: https://claude.com/blog/a-field-guide-to-claude-fable-finding-your-unknowns
• Digest slices: knowledge-corpus .work/youtube-watch/how-i-plan-build-and-run-loops-with-clau-aVO6E181cNU/ and field-guide-to-fable-thariq-shihipar-ant-9fubhllmsBU/ (recommendations/menu.md P0/P1 rows)

Pipeline

Brief (topic slice) → /discovery:explore → /planning:plan → implement, per the education plugin's conventions.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Y4RMR3G1rHAdNjxTqenVJL

[kyle-sexton]

Interview follow-up (2026-07-21) — retention requirement added:

• Primary value confirmed by operator use case: teammate answerability — "what did you do on ticket X (a week ago)" / "walk me through this feature." Quiz-me is retention-at-write vs the current fallback of retrieval-at-need (mining git history + ticket on demand).
• Retention seam (new requirement): quiz/report artifacts are NOT committed to the product repo (clutter). Configurable retention destination via userConfig; default a local disk library (knowledge-corpus is the natural existing home). Obsidian-vault adapter is a plausible future destination — record as deferred seam, do not design now.
• Retrieval flow (companion requirement): "what did we do on ticket X" should query the retained report library first, git/ticket archaeology second.
• Composes with the library_dir portability decision in knowledge: library_dir seam accepts only literal paths — machine-specific config collides with hardcoded-path guardrail #798 (same destination-config problem class).

🤖 Generated with Claude Code

https://claude.ai/code/session_01Y4RMR3G1rHAdNjxTqenVJL

[kyle-sexton]

[kyle-sexton]

Pipeline progress (session 2026-07-21): Brief + Tier-B design resolution + stress-tested five-phase plan committed on feat/799-education-quiz-me at docs/topics/education-quiz-me/PLAN.md (19373f7). Brief deferred questions resolved plan-side (name quiz-me grammar-checked; threshold >5 files / >200 LOC / blast ≥ HIGH; recall action; ${CLAUDE_PLUGIN_DATA} repo-identity-keyed retention default). Implementation blocked on plan approval.

[kyle-sexton]

This was generated by AI during triage.

Triage: briefed (delegable) → status: ready. Feature request; product design already locked by the operator's interview session (claude-loop-practices, 2026-07-21). Type Feature. Priority medium (education outside the week's priority-plugin set). Not a dup (education currently ships teach + explain; quiz-me is the third sibling). No open prerequisite.

Behavioral contract. Add an education-plugin skill that, after Claude completes a change, produces a report (HTML per the canonical pattern; markdown fallback per project convention) covering what was done — context, intuition, decisions — plus a quiz the user answers, verifying the human absorbed the work (representation accountability, loop retention, and secondary late intent-mismatch detection pre-merge).

Locked design (operator interview — do not re-open):

• The object under test is the human, not the artifact. Never overlaps verification:confirm (which owns artifact verification); compose by pointer only (e.g. an optional stage in a workflow gate list).
• Non-gating by default; opt-in gating via userConfig (e.g. quiz_policy: off | on-request | always | above-threshold, threshold keyed to change size / blast radius).

Decision defaulted (minor — veto before merge): the skill name — quiz-me is the working label; settle the final name via the codified naming grammar.

Pipeline: runs the standard Brief → discovery/explore → planning/plan → implement path; the planning stage is the design/approval gate for implementation specifics.