start_quiz
Play demoTry asking
- Start the Africa capitals quiz
- quiz me on African capitals
- start a new Africa quiz round
ghost skill run ghost/africa-quiz start_quiz Africa Quiz (F-Droid): a 10 question quiz on the capitals of African countries. Start a round, read and answer questions, then read the score and the answer review. Offline, no account.
Agent run: finished without the skill, finished with it (not independently checked).
One recorded run each way, claude-haiku-5-5, same phone and same prompt. Watch it in the Before / after tab.
1.1× fewer tokens, 3.1× slower with the skill. Agent run: finished without the skill, finished with it (not independently checked).
| Run | Time | Tokens | Tool calls | Result |
|---|---|---|---|---|
| No skill: agent drives by hand | 13.5 s | 33,503 | 4 | finished |
| With skill: find_skill, run_skill | 41.6 s | 30,051 | 4 | finished |
Prompt: Start the Africa capitals quiz, answer the first three questions with option 1 and tell me my score. Do not use the Play Store, no installs, no sign-in.
Same prompt, same phone, clean app state per side. Left: no skill, the agent drives the UI by hand. Right: find_skill + run_skill. 'ok' means the agent finished its turn; the result was not independently checked.
One recorded pair on samsung-a03s, claude-haiku-5-5; tokens count every input, cache and output token once per assistant message. Not a signed run: it adds no run evidence.
Control Africa Quiz on Android with an AI agent. Quiz yourself on African capitals: an agent starts a round, answers the questions and reads the review.
Target search phrase: AI agent plays a geography quiz on Android
Demo shows: answer (option=1)
Ghost in the Droid seed skill for Africa Quiz (
com.mufradat.africaquiz), an offline quiz on African capitals. Questions, options, the score and the answer review are plain text, so an agent can play and grade a whole round. Offline, no account.
| Action | What it does | Params |
|---|---|---|
start_quiz | Restart the app and start a new 10 question round. | none |
get_question | Read the current question: number, score, the prompt and the four options. | none |
answer | Pick an option by position (1-4) or by its text; the quiz moves on at once. | option, expect_question |
answer_many | Answer several questions in a row, e.g. options=1,3,2. | options |
get_results | After the last question: the percent, the score and the full answer review (country, correct capital, your answer). | none |
An agent can drive the whole app with these; compose them.
| Role | Action |
|---|---|
| Entry | start_quiz |
| Primitives | answer, answer_many |
| Read state | get_question, get_results |
The app: F-Droid, or https://f-droid.org/packages/com.mufradat.africaquiz/. The skill:
pip install "ghost-in-the-droid>=1.6.0" # 1.6.0 is not on PyPI yet: until it ships, git clone the repo and `pip install -e .`
ghost skill install ghost/africa-quiz
ghost skill run ghost/africa-quiz start_quiz --device <serial>
ghost skill run ghost/africa-quiz answer --param option=1 --device <serial>
run_skill(name="ghost/africa-quiz", action="answer", params={"option": "Kigali"})
# {"result": {"chosen": "Kigali", "state": {"number": 3, "score": 1}}}
Try asking: "Start the Africa capitals quiz, answer the first three questions and tell me my score."
Real Samsung phone, Android 13 (720x1600, 300 dpi, en-GB), Africa Quiz 1. License: MIT.
| App version | Android | Device | Skill version | Dates |
|---|---|---|---|---|
| 1 | 13 | Samsung real phone (720x1600, 300 dpi, en-GB) | 1.0.0 | 2026-11-02 |
SOFT.md is the narrative an agent reads when the exact steps no longer match your app version and it has to adapt the skill.
What each action means, which UI landmarks it relies on and how success is recognised, so an agent can adapt it to another app of this kind or another version.
Package: com.mufradat.africaquiz (Africa Quiz, MIT, version 1 from the official F-Droid repository, sha256-verified)
Verified: 2026-11-02 on a real Samsung phone, Android 13, 720x1600 @ 300 dpi, en-GB. Offline, no account.
Skill class: HARD. No login, no checkpoints.
Countdown Timer switch (15 seconds per question), Start Quiz.Question n/10, Score: s, the prompt as two texts (What is the capital of / Eritrea?), four option rows.Quiz Complete!, percent, k/10, then the Answer Review rows (Correct: ..., Your answer: ..., a tick or cross) and Play Again.start_quiz()Force-stops the app, relaunches and taps Start Quiz. The countdown timer switch keeps its default.
get_question()Reads Question n/10, Score, the prompt and the four option rows.
answer(option, expect_question)Taps the option row. The app has no per-answer feedback screen: read the score change or the final review.
answer_many(options)Reads the screen before every tap so the options of each question are looked up fresh.
get_results()Reads the result screen and scrolls to collect all 10 review rows.
The countdown switch is not read or changed (its state is not exposed).
Try a sample prompt or run an action by name. Replace any parameters in braces with your own values.
start_quizTry asking
ghost skill run ghost/africa-quiz start_quiz get_questionTry asking
ghost skill run ghost/africa-quiz get_question answerTry asking
ghost skill run ghost/africa-quiz answer answer_manyTry asking
ghost skill run ghost/africa-quiz answer_many get_resultsTry asking
ghost skill run ghost/africa-quiz get_results Each release is an immutable git tag. The tree hash pins the exact skill directory the installer verifies.
| Version | Released | App versions | Tree | Size | Provenance |
|---|---|---|---|---|---|
1.0.0latest | never | >=1.0 | pending | 21.4 KB | untagged |