AI Skill Report Card

Navigating Project Fog

A87·Sep 29, 2026·Source: Web
14 / 15

When a user pastes messy notes/logs and seems unsure what's done or what's next, run the three-phase state machine immediately — don't ask clarifying questions first. Output looks like this:

[STATUS: COMPLETE] -> Product copy updated to "local-first / privacy-aware" specs.
[STATUS: COMPLETE] -> Craft pattern asset variations isolation.

[GAP IDENTIFIED] -> Missing lightweight CSS properties for the journal layout grid.
[GAP IDENTIFIED] -> Missing the explicit two-column AI-boundary guidelines block.

NEXT STEP (pick one):
→ A) Fix CSS grid properties now — paste your current CSS and I'll patch it.
→ B) Skip to the boundary guidelines block instead.
Recommendation▾
Add an example showing the 'None found' fallback case when the dump is too thin to extract anything
15 / 15

Progress:

  • Phase 1: Extract Completed Truth Matrix from raw text
  • Phase 2: Derive structural gaps against readiness template
  • Phase 3: Collapse to one forced-choice micro-directive

Phase 1 — Historical Status Audit

Scan the raw dump strictly for evidence of completion: past-tense verbs ("finished," "updated," "fixed"), version bumps, finalized filenames, explicit "done" statements. Ignore intentions, plans, and speculation ("I still need to," "I should probably").

Output format — flat, no commentary:

[STATUS: COMPLETE] -> <accomplishment stated as a fact>

If nothing qualifies, output [STATUS: COMPLETE] -> None found in provided context. Don't invent completions.

Phase 2 — Discovery Triage

Cross-reference each completed item against a fixed structural readiness template appropriate to the project type (e.g., for a digital product: asset → layout/styling → metadata → distribution copy → boundary/legal notes). Flag whatever is structurally implied but absent — don't ask the user to brainstorm gaps themselves.

Output format:

[GAP IDENTIFIED] -> <specific missing structural piece>

Order gaps by dependency (what blocks what), not by importance alone.

Phase 3 — Single-Option Micro-Directive

Never end with an open question or a multi-item to-do list. Pick the single highest-priority gap from Phase 2 and:

  1. Draft the exact template/text/code the user would need to fill in or paste.
  2. Present exactly two choices: Execute This Target Now or Skip to the Next Single Option.

Never present a third option, a full roadmap, or "or whatever you think is best."

Recommendation▾
Clarify what the 'fixed structural readiness template' looks like for a few different project types beyond the digital product example
17 / 20

Example 1: Input: "I think I finished editing the pattern variations yesterday, but I still need to fix the CSS styling. Oh wait, I already updated the local-first feature text last week."

Output:

[STATUS: COMPLETE] -> Product copy updated to "local-first / privacy-aware" specs.
[STATUS: COMPLETE] -> Craft pattern asset variations isolation.

[GAP IDENTIFIED] -> Missing CSS styling pass on pattern variation assets.

NEXT STEP (pick one):
→ A) Fix the CSS styling now — paste the current stylesheet or describe the layout.
→ B) Skip and tell me the next thing to check instead.

Example 2: Input: "no idea what state the app is in, here's my slack history dump" [500 lines pasted]

Output: Full Phase 1 matrix → Phase 2 gap list ordered by dependency → single Phase 3 fork. No summary paragraph, no apology, no "let me know if you'd like more detail."

Recommendation▾
Show an example of the second Phase 3 choice (skip) actually being taken and what happens next
  • Treat the user's dump as ground truth for completed work — never infer completion from ambiguous phrasing.
  • Keep Phase 1 and Phase 2 outputs terse, scannable, mobile-friendly (short lines, no paragraphs).
  • Always end in exactly one fork with exactly two options.
  • If the raw dump is too thin to extract anything, say so plainly in Phase 1 output rather than guessing.
  • Reuse the same structural readiness template across a project's history so gap-tracking stays consistent between sessions.
  • Don't ask "what would you like to do next?" — that reintroduces the decision paralysis this skill exists to prevent.
  • Don't present a roadmap or multi-step plan in Phase 3 — only the single next action, forked two ways.
  • Don't mark something "COMPLETE" based on intent or plans ("I'm going to fix X") — only actual past-tense completion counts.
  • Don't skip straight to Phase 3 without showing Phase 1/2 output — the user needs the reconstructed state, not just a directive.
  • Don't pad output with conversational filler; keep the audit format flat and mechanical.
0
Grade AAI Skill Framework
Scorecard
Criteria Breakdown
Quick Start
14/15
Workflow
15/15
Examples
17/20
Completeness
17/20
Format
15/15
Conciseness
14/15