← Back to T1ENS5 · Forge — Raise Your Wukong Well
T1ENS5 · Course material

✏️ S5 Homework Guide (submit · Golden Hoop 5 · pick one cut, change it, run it)

S5 Homework (submit · Golden Hoop 5 · Refinement) — Forge Your Little Wukong · Change One Real Cut

This is the student playbook for the S5 required homework.

The real point of this homework: in S4 you wrote a refined script. S5 is not "write another version" — S5 is picking one place, actually changing it, and running it once to see the effect.

An agent is not a document — it is forged.

Deadline: Sat 8/29 · 9:00 AM PT (before S6 begins) · AI-graded

Submit at: https://www.siliconroshi.ai/homework (after login, auto-redirects to /homework/mine)

Course ID: ENS5-FORGE


The one thing you submit (in one sentence)

Pick the one place in your S4 script that most needs changing, actually change it on WukongDojo (edit the script or feed the KB), run it once to see the effect, and write the result down.


The homework in 2 steps

┌─────────────────────┐   ┌─────────────────────┐
│ Step 1              │ → │ Step 2              │
│                     │   │                     │
│ Pick one cut +      │   │ Run it once +       │
│ actually change it  │   │ write it down       │
│                     │   │                     │
│ edit script or KB   │   │ see effect · log it │
│ (30 min)            │   │ (30 min)            │
└─────────────────────┘   └─────────────────────┘

Total time: about 1 hour.


Step 1 · Pick one cut + actually change it

Open your S4 refined script and find the one place that most needs changing.

It could be:

  • A step whose leading strategy is written too vaguely (edit script)
  • A pushback you KNOW customers will raise but the agent can't answer (feed KB)
  • A closing criterion that's too fuzzy (edit script)
  • An OUT-OF-SCOPE boundary that isn't held (feed KB)

⚔️ Hard constraint: pick only one cut. Forge one deep >> graze seven shallow.

Two ways to change it (do at least one)

Way A · Edit the script

Edit your agent's script directly on WukongDojo. When done, you must be able to say which line you changed, and from what to what.

Way B · Feed the KB

Take one of the .md files you listed in S4 step 4 and actually write it out — at least 3 concrete entries — and feed it into WukongDojo's KB.

💡 About Memory: WukongDojo's Memory is still on the roadmap; you don't need it for S5. Editing the script + feeding the KB is already enough to forge one real difference.


Step 2 · Run it once + write a forge log

After changing it, think of a real scenario (not a test question), paste it in, run it once, and see the effect.

Then write a forge log with the following three sections:

Section 1 · What I changed

My agent category: ______ (carry the A / B / C you declared in S4)
The one cut I changed: ______________________________
Method: □ edit script  □ feed KB  □ did both

Concrete change:
  (if editing script) BEFORE: "____________" → AFTER: "____________"
  (if feeding KB) filename + entries:
    1. ____________
    2. ____________
    3. ____________

Section 2 · How it went

Scenario: ____________________________________________________

Dialog excerpt (pick the 2-3 most interesting lines of the agent):
  agent: ____________________________________________
  me (as customer): ___________________________________
  agent: ____________________________________________

Effect: ____________ (clearly better / about the same / actually worse)
Why in one sentence: ______________________________________

Section 3 · Next step

What I still want to change before S6: ______________________________
One specific question for Bill: ________________________

📐 Category hints

Depending on the category you declared in S4, the leverage of your polish differs:

Category Where it's most worth forging
A Closer Which step's closing criterion is weakest / where customers stall most
B Q&A Which question type's OUT-OF-SCOPE is easiest to blow
C Companion Which move you trust least (placating / missing ESCALATION)

⚠️ A few reminders

  • You must be able to name the concrete change. "I polished it a bit on WukongDojo" without being able to say what you changed = no polish.
  • Running it once is required. If you don't run it, you don't know if the change landed.
  • Write the effect honestly. If it got worse, that's a valuable observation — you learned the cost of constraints.
  • Feeding the KB means pasting content. Filename without concrete entries = not actually written yet.

📋 Pre-submit checklist

[ ] Wrote the category (A/B/C)
[ ] Named which one cut you changed
[ ] Concrete change (before → after, or at least 3 KB entries)
[ ] Ran it once, pasted 2-3 dialog lines
[ ] Wrote how it went + why
[ ] Wrote what to change next

How to submit

Submit at: https://www.siliconroshi.ai/homework

Paste the three sections above:

  • Section 1 · What I changed
  • Section 2 · How it went
  • Section 3 · Next step

Course ID: ENS5-FORGE


⚠️ On "letting AI finish the homework"

This S5 homework doesn't survive AI ghost-writing — because what it asks for is the live record of you actually forging that one cut with your own Little Wukong. Which cut you were least sure of, whether the agent was truly different after the change — AI doesn't have that; only you, who actually forged it this week, do.

In the S6 class we'll take this log and walk it toward launch.

—— An agent is not a document — it is forged.


🎁 Appendix · Bonus: run it past a real person (optional)

If your polish already holds up, reward yourself by running it past a real person:

Who: ______________
What they said after using it: ________________________________
What I saw: ______________________________________

Write it at the end of Section 3. Skipping it costs no points.


📚 Appendix 2 · Example log (study the structure, don't copy the content)

Section 1 · What I changed
  Category: A Closer
  The cut: Step 3 leading strategy — "customer admits the 30-year gap is real"
  Method: did both

  Edit script:
    BEFORE: "before I run the numbers, let me explain the compounding formula"
    AFTER:  "before I run the numbers · what do you guess the 30-year multiple is?"

  Feed KB: pushback-bank
    1. "7% is optimistic" → give three brackets 5%/7%/9%, let the customer pick one
    2. "insurance compounds too" → cite the IUL cap + real cost breakdown
    3. "I have no money to invest now" → don't push, ask "what about 30 years from
       now if you leave this money alone"

Section 2 · How it went
  Scenario: "my wife asked me last night what to do with $300K sitting 18 months
             in a 0.4% savings account"

  Dialog excerpt:
    agent: before I run the numbers · what do you guess the 30-year multiple is?
    me (as customer): hmm… maybe 2x?
    agent: real data at 7% over 30 years: $300K → $2.28M (7.6x). You say 7% is
           optimistic? Let me plot all three — 5/7/9 — the 5% gap is still $1.3M.
           Want to keep going at 5%?

  Effect: clearly better.
  Why: before, the agent dropped its standard whenever pushed back (it caved); now
       it offers three brackets and lets the customer pick, without losing the
       sub-goal. The pushback-bank standard re-lead worked.

Section 3 · Next step
  Still want to change: add a few more pushback-bank entries (only wrote 3 this time)
  Question for Bill: how many pushback-bank entries are usually enough to support a
                     launch?

One last reminder: S5 is not writing a report — it's actually having forged that one cut with your own Little Wukong.

Forge that one cut, write it down — S6 we walk it toward launch.