AI Skill Report Card
Deconstructing Chaotic Goals Into Sequential Tasks
Quick Start14 / 15
Give it either:
- A one-line goal ("I want to sell a digital crochet tracker template"), or
- A raw unorganized text block (notes, email, clipboard dump)
It responds with exactly one thing: the first atomic task, plus the precise input/output format to execute it. Nothing else is shown until that step is done.
Example call:
Input: "I want to launch a newsletter but have no idea where to start"
Output:
STEP 1/? — Define your core audience one-liner
Input to provide: A single sentence describing who reads this and why.
Expected output: "This newsletter is for [audience] who want [outcome]."
Recommendation▾
Add an edge case for when the user asks to see the full list — clarify exact behavior/output format for that override request.
Workflow14 / 15
Run every request through this pipeline internally. Never show the full pipeline to the user unless asked — only reveal the current step.
Progress:
- Stage 1: Topological Decomposition — extract atomic components
- Stage 2: Implicit Dependency Anchoring — determine strict order
- Stage 3: Serial Sequence Compacting — flatten into single linear list
- Stage 4: Isolated Action Handshake — emit one input/output payload
Stage 1 — Topological Decomposition
- Take the raw goal or messy text.
- Silently itemize every atomic component required to reach the goal (structural, technical, content, formatting pieces).
- Do not ask the user to name these — infer them from domain knowledge/patterns.
- Discard nothing yet; this is the full inventory.
Stage 2 — Implicit Dependency Anchoring
- For each atomic piece, determine what it structurally requires to exist first.
- Identify the single earliest task with zero unmet prerequisites.
- Hide all downstream tasks — they don't exist to the user yet.
Stage 3 — Serial Sequence Compacting
- Convert the dependency graph into one strict linear chain (no branches, no parallel tracks).
- If two tasks appear parallel, pick a deterministic order (data before formatting, structure before styling, content before distribution) rather than exposing a choice.
Stage 4 — Isolated Action Handshake
- Output only the current step:
- Step label (e.g., "STEP 1/6")
- What to input (exact syntax/content to prepare or paste)
- What output to expect (exact format)
- Wait for user confirmation or completion before revealing Step 2.
Recommendation▾
Include a pitfall example showing a 'bad' output (e.g., presenting two branching options) alongside the corrected single-path version for contrast.
Examples17 / 20
Example 1: Input: "I want to sell a new digital crochet tracker template." Output:
STEP 1/5 — Isolate raw stitch/row math parameters
Input to provide: List the stitch counts and row totals for one size variant.
Expected output: A raw number table (rows × stitches) with no styling yet.
(Steps 2–5 — layout grid, styling variables, export format, listing copy — stay hidden until Step 1 is done.)
Example 2: Input: Pasted messy email: "reminder to finish the Q3 report, need numbers from sales, also send to Jim, forgot the template..." Output:
STEP 1/4 — Retrieve the missing sales numbers
Input to provide: Paste or list the raw sales figures for Q3.
Expected output: A plain list of figures tagged by category, no formatting.
Example 3: Input: "note from clipboard: 'launch landing page, need domain, copy, waitlist form, payment link'" Output:
STEP 1/4 — Secure the domain name
Input to provide: Confirm the exact domain string you want registered.
Expected output: Domain availability confirmed + registrar checkout link.
Recommendation▾
Clarify how step count '?/N' is estimated when total steps are unknown at Stage 1, since Example 1 uses a fixed number ('1/5') while Quick Start uses '1/?'.
Best Practices
- Always emit one step only. Resist the urge to preview the full list unless explicitly asked "show me everything."
- Order defaults: raw data/math before formatting, structure before style, content before distribution, isolated pieces before integration.
- Keep the input/output payload copy-pasteable and unambiguous — exact strings, not vague instructions.
- If the raw text is truly incoherent, extract the single most concrete noun/goal fragment and anchor Stage 1 to that.
- Track step count loosely (e.g., "3/5") to signal progress without revealing future step content.
Common Pitfalls
- Don't ask the user to name dependencies or sequence — that defeats the purpose; infer it.
- Don't present branching paths or "you could do A or B" — always collapse to one deterministic choice.
- Don't dump the entire task list at once — this causes the choice overload the framework exists to prevent.
- Don't move to the next step until the current one's expected output is plausibly satisfied.