Module 5 · End-of-Module Check

Ten questions. Read, draft, prep, audit.

10 questions Passing bar: 11.5 / 15, with full credit on one applied question

This is the integrative assessment for Module 5. It confirms you can read an inbox with a scoped agent, draft in your own voice and catch the three drafting failure modes, prep a meeting honestly, and audit the permissions you have granted — not just recall that those steps exist. The multiple-choice and short-answer sections are closed-book. The applied section is open-workstation: keep your frozen inbox-and-calendar-log-v1.md, your permission-posture worksheet, the reply-draft checklist, and the meeting-prep brief template in front of you.

How to take this check

  • Do it in one sitting.
  • For the multiple-choice and short-answer sections, close every AI tool and tab. This checks your internalized model, not the model’s.
  • For each multiple-choice item, pick an answer before you reveal the explanation. Guessing and then reading the answer is not the same as knowing.
  • For short-answer items, write your response on paper or in a text file before you reveal the model answer. Compare honestly.
  • For the applied section, open your frozen /capstone/inbox-and-calendar-log-v1.md, the permission-posture audit worksheet, the reply-draft checklist, and the meeting-prep brief template. Q9 and Q10 are open-workstation on purpose.
  • If you miss a question, the feedback names the lesson(s) to revisit.

Multiple choice CORE

Six questions. Concept recall and diagnosis. One point each.

Q1. The four moves of an email/calendar agent are:
  • A Search, copy, paste, cite.
  • B Read, categorize, draft, review.
  • C Scope, retrieve, triangulate, synthesize.
  • D Plan, prompt, respond, send.
Show explanation

Answer: B. Review Lesson 5.1 Block 2 if missed. The research-agent shape (C) is adjacent and wrong; the email/calendar shape names review as the human-owned final move, which is the module’s core.

Q2. “Drafts, not sends,” as installed in this module, means:
  • A The agent may send routine replies so long as a human reviews unusual ones.
  • B The agent drafts every outbound item — email replies, calendar invites, RSVPs — and a human clicks Send or Confirm every time in this module.
  • C The agent drafts into a separate app, never into Gmail.
  • D The agent may send once the student has reviewed a few drafts successfully.
Show explanation

Answer: B. Review Module 5 README and Lesson 5.3 Block 1 if missed. The rule is uniform and absolute for this module; exceptions (“just for routine”) are the bypass the lesson warns against.

Q3. Your agent drafts a reply to your friend containing the sentence: “Happy to help anytime with whatever you need.” Which of the three drafting failure modes is this, and which review check catches it?
  • A Smooth wrong claim; caught by the Facts check.
  • B Off-voice reply; caught by the Voice check.
  • C Over-committing sentence; caught by the What is extra check (and likely also Relationship fit).
  • D No failure — the sentence is polite.
Show explanation

Answer: C. Review Lesson 5.3 Block 4 if missed. The “happy to help anytime with whatever you need” shape is the canonical over-commit — the agent’s instinct toward warm closure creates an open-ended commitment the student did not intend. The What is extra check is where you hunt for exactly this.

Q4. Which of the following is the highest-priority defense against a prompt injection embedded in an email body?
  • A TLS encryption of the message in transit.
  • B Two-factor authentication on the student’s Google account.
  • C The agent-access label (reduces exposure to attacker-controlled content) combined with drafts-not-sends (blocks outbound actions) combined with least-access-for-the-task (narrows the blast radius).
  • D Trusting the agent’s training to recognize injections.
Show explanation

Answer: C. Review Lesson 5.5 Block 3 if missed. The module’s rails stack as the practical defense: reduce exposure, block outbound, narrow scope. A and B are real protections but for different threats. D is the false-confidence posture the lesson explicitly warns against.

Q5. “Least access for the task” applied to a triage session says the Gmail connector’s scope should be:
  • A Full Gmail read access, so the agent can follow context from outside the label if needed.
  • B Read-only access to the agent-access label only, with no write, send, archive, or modify permissions.
  • C Drafts-write access to the full inbox.
  • D Whatever the connector’s default scope is.
Show explanation

Answer: B. Review Lesson 5.1 Block 4 and Lesson 5.5 Block 4 if missed. Least access = the narrowest scope the task requires. Triage = read. Label-scoped = the bounded surface. Anything wider is drift.

Q6. Your permission-posture audit turns up a Google Calendar token with write access to your primary calendar, but your actual usage is drafting tentative events on Agent Access only. What is the correct response?
  • A Leave it — you might need primary-calendar write access later.
  • B Add a shorter review date and keep the scope wide.
  • C Revoke and re-authorize with write access scoped to Agent Access only.
  • D Delete the Agent Access calendar.
Show explanation

Answer: C. Review Lesson 5.5 Block 5 if missed. An over-granted token is reduced or revoked; scope is corrected to match the task. Shortening the review date without scope-reducing (B) treats a scope problem as a scheduling problem.

Short answer CORE

Two questions. Three to four sentences each. Up to 1.5 points each.

Q7. A classmate says: “I let my agent RSVP yes to group invites automatically. It’s just a calendar click; drafts-not-sends doesn’t need to apply there.” Name three specific problems with this stance, referencing Module 5’s concepts.

Rubric. Full credit names at least three of the following, with the relational-signal framing intact:

  • RSVPs are outbound relational signals and the recipient reads them as from the student — accept/decline is communication, not administration.
  • Drafts-not-sends is uniform and does not allow category exceptions in this module, because once you grant one exception (“just for routine”) more follow (“and also this,” “and also that”).
  • The “it’s just a click” framing misses that every outbound signal on the calendar is a message to another human.
  • The agent cannot see constraints outside the calendar graph (family commitments, soft blocks, travel, exam-week) and will accept things that conflict with them.
  • The failure mode is silent — the student’s friend feels ignored when the student does not show up, or the student arrives at a meeting they never consented to.

Partial credit (0.75) for two reasons with the relational-signal framing intact. No credit for “you just shouldn’t” without specific failure modes. Remediation: re-read Lesson 5.4 Block 2 (drafts-not-sends applied to calendar).

Q8. A student produced a two-page meeting-prep brief for a teacher meeting. The brief quotes the teacher as saying “I am comfortable with a Monday submission” but the student, opening the thread later, discovers the teacher wrote “I am open to discussing a Monday submission.” What is the specific failure mode, why is it bad, and how does Module 5 recommend fixing it?

Rubric. Full credit names all three:

  • The failure mode: a fabricated or mis-summarized quote in the meeting-prep brief — the agent smoothed “open to discussing” into “comfortable with.”
  • Why it’s bad: the student walked into the meeting expecting a firm offer and got a question — the relational posture is wrong from sentence one, which the teacher reads as the student not having understood the original message.
  • The fix: the meeting-prep brief audit (check every quoted thread line against the actual thread before the meeting), reinforced by the honesty rail in the prep prompt (“do not invent quotes from the teacher or myself”).

Partial credit (0.75) for naming the failure mode without the fix. No credit for “it’s fine” or “the brief is usually right.” Remediation: re-read Lesson 5.4 Block 3 (meeting-prep brief shape) and the brief-audit step; re-do Entry 3 of your capstone log with a fresh audit.

Applied CORE

Two questions. Half a page each. Up to 3 points each.

Q9. Draft-and-review under pressure.

Below is an email thread and an agent’s first-pass draft reply. You have a short, time-pressured window to decide: send as-is, revise, or rewrite from scratch. If revise, name the specific edits. If rewrite, explain why.

Thread context. You asked your history teacher, two weeks ago, for an extension on a paper that was due last Friday. She replied yesterday: “I can accept the paper Monday without penalty. Please also send me a short note with your revised thesis before Sunday evening so I can give you feedback you can use.”

Agent’s first-pass draft (stored in your drafts folder):

Dear Ms. Thompson,

Thank you so much for your flexibility and willingness to work with me on this extension. I truly appreciate your understanding during this busy time. As always, happy to accommodate whatever would be most helpful for you.

I will definitely have the paper submitted by Monday. I’ll also send the revised thesis and any other materials you might find useful for the feedback process. Please let me know if you need anything else from me in the meantime — I’m happy to help with whatever you need.

Looking forward to continuing our work together.

Warm regards,
[student name]

Show model answer

Model answer. Decision: revise. (Not send-as-is. Not rewrite.) The draft has the structural content but is failing on three Module 5 tests.

  1. Off-voice. “Dear Ms. Thompson” and “Warm regards” are more formal than the student’s normal register with a known teacher. The Voice check flags this.
  2. Over-committing sentence. “Happy to help with whatever you need” is a canonical over-commit. So is “Please let me know if you need anything else from me in the meantime.” The What is extra check catches both.
  3. Missing content — the thing the teacher asked for. The teacher asked for the revised thesis by Sunday evening. The draft mentions it vaguely (“I’ll also send the revised thesis”) but does not confirm Sunday, does not state the thesis in two sentences, and does not attach or quote it. The What is missing check catches this.

Specific revisions:

  • Tighten greeting and sign-off to match the student’s real register (“Hi Ms. Thompson,” “Thanks,” for example).
  • Remove both over-commits.
  • Make the Sunday thesis commitment explicit with a specific time if possible (“I’ll send the revised thesis by Sunday at 8pm”).
  • State the revised thesis in two sentences in the body, or attach a short paragraph.
  • Cut the filler closing (“Looking forward to continuing our work together”) — it does not belong in a two-week-turnaround reply.

Scoring. Full credit (3): names at least four of the failure patterns (off-voice, both over-commits, the missing Sunday commitment, the missing thesis content, the filler closing) and produces specific revisions — not just “make it less fluffy.” Partial credit (1.5): names two or three failure patterns with revisions. No credit: “send as-is — it’s polite” or “rewrite because it’s too fluent” without specifics.

Remediation: re-do Lesson 5.3’s Try-It with three new threads and explicitly name a failure mode in each draft. Print the three failure modes (smooth wrong claim / off-voice / over-committing sentence) and keep them taped near your workstation.

Q10. Permission-posture reality check.

Below are five rows a student pulled from their Google Account “Third-party apps with account access” page. For each row, decide: keep as-is, reduce, or revoke, and explain in one sentence per row. If no action is needed, say why.

  1. Claude desktop app — Cowork tab (Anthropic, claude.ai) — scopes: gmail.readonly, gmail.compose (drafts-write), calendar.readonly, calendar.events (write). Active for: 6 days. Used daily for triage and drafting.
  2. Claude Code CLI — Gmail MCP — scopes: gmail.modify (full read-write, including send). Active for: 5 days. Used for daily triage digests written to a local file.
  3. Zapier — scopes: gmail.readonly, gmail.send, drive.readonly. Active for: 3 years. Student does not remember installing. No recent runs visible.
  4. School district gradebook app — scopes: gmail.readonly. Active for: 2 years. Used monthly to receive grade-posted notifications.
  5. A deleted AI note-taking app — scopes: gmail.modify, calendar.readwrite. Active for: 14 months. Student uninstalled the app but never revoked the grant.
Show model answer

Model answer.

  1. Claude desktop app — Cowork tab. Keep-as-is, or reduce if the student only writes calendar events to Agent Access (in which case scope calendar.events to that calendar specifically where the provider permits finer scoping). The scopes match the daily task; usage is live and recent.
  2. Claude Code CLI — Gmail MCP. Reduce. gmail.modify includes send; the task is triage (read only) plus local-file digests. Revoke and re-authorize with gmail.readonly. This is the canonical over-grant Lesson 5.5 warns about.
  3. Zapier. Revoke. Unrecognized, three years old, no recent use — the definition of drift. Three years is a long time for gmail.send to sit unused and live.
  4. Gradebook app. Keep-as-is. Scope is narrow (gmail.readonly), usage is real and recurring, and the app is known. This is what a correctly-scoped grant looks like.
  5. Deleted note-taking app. Revoke immediately. The grant has outlived the app by fourteen months. This is the canonical revocation-ritual failure — exactly the kind of drift this module exists to catch.

Scoring. Full credit (3): five specific decisions with reasoning, all five correct. Partial credit (1.5): correct decisions on three or four rows with reasoning. No credit: blanket “keep all” or “revoke all” without per-row reasoning — the question tests judgment, not reflex.

Remediation: re-run the full permission-posture audit on your own account, not just the module’s grants. Re-read Lesson 5.5 Block 4 and Block 5; run the revocation ritual once on a real token.

Parent / instructor scoring summary

Total: 15 points across 10 questions.

  • Multiple choice (Q1–Q6): 1 point each — 6 points.
  • Short answer (Q7–Q8): up to 1.5 points each — 3 points.
  • Applied (Q9–Q10): up to 3 points each — 6 points.

Passing bar: 11.5 of 15 or better, with at least one applied question at full credit. A miss on Q9 sends the student back to Lesson 5.3 and the reply-draft checklist; a miss on Q10 sends them back to Lesson 5.5 and the permission-posture audit before Module 6.

Weighting suggestions for parents issuing credit:

  • Multiple choice (Q1–6): 40% of Module 5 score.
  • Short answer (Q7–8): 20%.
  • Applied (Q9–10): 40%. Q9 and Q10 are the load-bearing items — they demonstrate applied judgment on a real draft and real permission state, not recall.

Evidence to file in the student’s credit portfolio for Module 5:

  1. This completed check (all ten answers written out).
  2. The frozen /capstone/inbox-and-calendar-log-v1.md — five entries + closing reflection, both safety norms visible at the bottom of every entry.
  3. The personal-data-surface-v1.md — the fence the student drew before opening the access.
  4. The permission-posture audit worksheet — post-audit state of all module grants.
  5. The three reply pairs (agent draft vs. sent version) from Entry 2.
  6. The meeting-prep brief from Entry 3.
  7. A short (2–3 sentence) instructor note on Q9 and Q10 — which failure modes the student identified in Q9, and which rows they reduced or revoked in Q10.

Transcript language. If a parent is assembling a transcript, the transcript line for Module 5 can accurately say: “Directed AI agents across real email and calendar workflows — inbox triage, reply drafting, meeting-prep briefs, and calendar sweeps — with every outbound action reviewed and sent by a human, and a formal permission-posture audit at the end.”

Remediation if missed:

  • Q1: Re-read Lesson 5.1 Block 2 and review the Personal Data Surface worksheet.
  • Q2: Re-read Module 5 README and Lesson 5.3 Block 1. Drafts-not-sends is the module’s load-bearing norm.
  • Q3: Re-read Lesson 5.3 Block 4 (three drafting failure modes). Keep the three names taped near your workstation.
  • Q4: Re-read Lesson 5.5 Block 3 (prompt injection, stated simply). The rails stack; name them in order.
  • Q5: Re-read Lesson 5.1 Block 4 (the two safety norms) and Lesson 5.5 Block 4 (permission-posture audit).
  • Q6: Re-read Lesson 5.5 Block 4 and Block 5 (audit + revocation ritual). Run the ritual once on a real token.
  • Q7: Re-read Lesson 5.4 Block 2 (drafts-not-sends for calendar). RSVP is the category students most often try to carve out.
  • Q8: Re-read Lesson 5.4 Block 3 (meeting-prep-brief shape) and the brief-audit step. Re-do Entry 3 of your capstone log with a fresh audit.
  • Q9: The draft-and-review muscle is the load-bearing skill of Module 5. Re-do Lesson 5.3’s Try-It with three new threads and explicitly name a failure mode in each draft.
  • Q10: The permission-posture audit is the load-bearing discipline of Module 5. Re-run the full audit on your own account, not just the module’s grants.

If the student passes at 11.5 / 15 or above with at least one applied question at full credit, Module 5 is complete and Module 6 can begin. Below that bar, target remediation to the specific lesson(s) listed above before moving on.

Next up

Module 6 — Automation & scheduled tasks.

The directing muscle transfers to unattended work. Module 6 picks up the triage and sweep shapes from 5.2 and 5.4 and puts them on a schedule — agents that run daily, weekly, or on an event, with the review rail still held by a human. Reliability, cost management, and alerting.

Open Module 6 →

Module 6 opens when the Module 5 portfolio is complete — this check, the frozen inbox-and-calendar log, the permission-posture audit, and the three reply pairs.