Darasa Hub - Learning Lions
Dira Portal Safety and Welfare ICT Support
FS5

Working with AI

Level 2 · BootCampFS FoundationsFS5a: Taught week 1 Thursday morning

Start the Lesson

By the end you can run the five moves on a real task of your own, spec it before the tool sees anything, catch a wrong line against a witness that is not the AI, label the work and keep the receipts, send one task to Claude, ChatGPT and Gemini and say from a card which did it best and at what cost, say what a token, a context window and an agent are, write a brief and a rules file that a tool reads, and watch an agent run without letting a single step past unread.

Start the lesson 77 moves in 14 parts. You need: your FS2 install notes sheet with a real question on it · your notebook and a pen · a programme account on the AI tools, by the access route your SA gives you · your copy of the programme folder, from FS4 · for the second block, the Director log from the first

Submit Your Work

What You Turn In

  1. Your Director log (FS5a) fill in on the hub
    1. A real question of your own; your attempt written before the spec; a spec with goal, context, limits, what good looks like; one line changed or rejected with the reason; every kept line with its outside witness; the label; the receipts; the honesty test passed
  2. Your second Director log fill in on the hub
    1. A different task from the afternoon SGL block, written up clean as page two; a real catch with its witness; by Saturday's update
  3. Your three-tool comparison card fill in on the hub
    1. The same task and spec to Claude, ChatGPT and Gemini; what each produced, one thing each got wrong or 'nothing found, checked against …', the tier, your verdict with a reason
  4. Your standing-instructions brief, rules file and prompt library
    1. A brief in four parts put into each tool and tested on a second task; a short rules file at the top of the programme folder; the library grown to five, each with a when-to-use line
  5. One skill and the token-budget line
    1. A skill made from a library entry and run once; two token counts on the same task, the long chat and the fresh chat with a context file
  6. Your agent-run log, with receipts fill in on the hub
    1. Every step as it ran: what it did, what you asked, what you did not ask, accepted or not; at least one honest 'did not ask' line; nothing private, no money, no credentials

Stuck, or want to check your thinking? Paste this card into your own AI and ask it to question you, not answer for you.

How

  1. Asha: FS5a in Thursday afternoon's Asha block; FS5b by the Saturday update of week 5.
  2. Free AI Tools: what each is best for and how much you get free. In the BootCamp folder on Mufasa.
Done when
  1. Director log, FS5a: Real own task; attempt first; four-part spec; one line changed or rejected with reason; every kept line with an outside witness; label; honesty test passed
  2. Second Director log: A different real task; a real catch with its witness
  3. Three-tool comparison card: Three tools, same task; one error each or a checked 'nothing found'; tier; verdict with a reason
  4. Standing-instructions brief + rules file + prompt library: Four-part brief put into each tool and tested; short rules file in the programme folder; library at five
  5. Skill + token-budget line: One skill made from a library entry and run; two token counts on the same task
  6. Agent-run log: Every step logged as it ran; a 'did not ask' line or a scope note; nothing forbidden handed over

Counts toward: the FS5a and FS5b badges, 1 to 3 stars each, the habit not the answer · the label and receipts on every badge from here · attendance for the Taught and SGL blocks · LP (Lion Points) for accepted badge work.

Fill in the worksheet Submit work for FS5