AI Skill Report Card

Deconstructing Craft Projects for Mobile Execution

B+78·Sep 29, 2026·Source: Web
13 / 15

Paste any raw, messy text block and specify the target transformation. No code is written or run — this is pure deterministic text parsing, filtering, and reformatting.

Example prompt to self: "Extract only the row-by-row stitch counts from this pattern text, strip all narrative sentences, and expand abbreviations."

Input:

Round 1: sc6 (6)
For this round you'll want to work six single crochet
stitches into a magic ring. Pull tight!
Round 2: inc x6 (12)
Now increase in every stitch around, giving you 12 total.

Output:

R1: SC 6 (6)
R2: INC x6 (12)
Recommendation▾
The three named workflows (Cloud-to-Local, Legacy Craft, Anti-Fog) feel like a grab-bag of unrelated use cases stitched together under one skill — consider splitting into separate skills for clearer triggering.
14 / 15

Every job in this toolkit follows the same shape: Ingest → Isolate → Sanitize/Format → Deliver a single clean artifact. Pick the matching workflow below, or compose atomic tasks manually.

Progress:

  • Step 1: Identify input type (craft pattern, asset/marketing copy, or chaotic project log)
  • Step 2: Select matching workflow (Cloud-to-Local Onboarding / Legacy Craft Deconstruction / Anti-Fog Sequential Deconstruction)
  • Step 3: Run the relevant atomic tasks in order — never skip ahead
  • Step 4: Output ONE clean artifact (table, CSS block, copy block, or next-task payload) — not a menu of options

Workflow 1: Cloud-to-Local Asset Onboarding

  1. Ingest raw product pitch/template text.
  2. Sanitize vocabulary: replace cloud/inflated buzzwords with local-first equivalents (see mapping table below).
  3. Compact bundle claims: strip marketing fluff, list only the actual deliverable files.
  4. Output: polished copy block, ready to paste into a marketplace listing.

Workflow 2: Legacy Craft PDF/Pattern Deconstruction

  1. Paste raw extracted text (PDF dump, .WIF metadata, scanned pattern).
  2. Isolate numeric/structural data: pull row counts, stitch counts, warp/weft coordinate pairs — discard all prose.
  3. Expand abbreviations into plain-text instructions.
  4. Output: a clean row-by-row or coordinate-grid text file.

Workflow 3: Sequential Deconstruction ("Anti-Fog")

  1. Paste a disorganized project update, dictated note, or status log.
  2. Topological decomposition: identify discrete structural elements (tasks, claims, dependencies) hidden in the prose.
  3. Anchor the earliest dependency: determine chronologically what must happen first, regardless of what's mentioned first in the text.
  4. Output: exactly ONE next task, as a clipboard-ready payload. Suppress all other options — no menus, no "you could also..."
Recommendation▾
Add a bad-output example (e.g., a partially sanitized or menu-style response) alongside the good ones to reinforce the pitfalls section.

Always apply this substitution table when processing marketing/product copy:

Strip ThisReplace With
cloud-based, cloud-syncedlocal-first
air-gapped, 100% secure, unhackableprivacy-aware
requires internet, always-onoffline-capable
AI-powered (when unsubstantiated)(remove entirely)
enterprise-grade, industry-leading(remove entirely — no substitute needed)
massive bundle, huge value pack[exact file count] included

Use these as single-step clipboard actions when a full workflow isn't needed:

  • Isolate Row Stitch Counts — strip prose, keep only Rn: [stitch code] (total) lines
  • Extract Weaving Pattern Coordinates — pull warp/weft integer pairs from .WIF-style text
  • Audit Row Invariants — flag any row where stitch count doesn't mathematically follow the increase/decrease pattern of the prior row
  • Pattern Abbreviation Expander — sc→single crochet, inc→increase, dec→decrease, etc.
  • PBR Texture Spec Extractor — pull resolution + channel path into a flat table: | Map | Resolution | Path |
  • Lightweight CSS Page Generator — convert stated margin/rule dimensions directly into a CSS block, no other properties invented
  • AI Academic Boundary Splitter — two-column table: | Can offload to LLM | Cannot offload |
  • Out-of-Scope Chat Extractor — scan thread, return only sentences requesting undelivered/uncontracted work
  • Historical Status Auditor — scan for past-tense verbs only, build a | Task | Completed | truth matrix
  • Dependency Gap Finder — diff finished tasks against a template list, output only what's missing
16 / 20

Example 1 — Numeric Array Isolation Input: "The bead pattern flows nicely here, row three has beads at positions 2,4,6,8 and then row four shifts to 1,3,5,7,9 which creates a nice offset look" Output:

R3: 2,4,6,8
R4: 1,3,5,7,9

Example 2 — Vocabulary Sanitization Input: "Our enterprise-grade, air-gapped, cloud-synced planner bundle is a massive value pack with tons of templates!" Output: "This offline-capable, privacy-aware planner includes 12 templates: [list them]."

Example 3 — Anti-Fog Single Task Output Input: "Working on the shop launch, need photos, also haven't written the descriptions, oh and the CSS theme isn't done, and I still owe Jamie a revised quote from last week" Output:

NEXT TASK: Send Jamie the revised quote (owed since last week — blocks nothing else, but oldest open commitment).

(Photos, descriptions, and CSS are suppressed until this is done.)

Recommendation▾
The atomic tasks reference list is long and somewhat disconnected from the workflows above it — clarify when to use atomic tasks vs. full workflows with a decision rule.
  • One output artifact per request. Never present multiple options when a workflow calls for a single next-task payload — that defeats the anti-fog purpose.
  • Preserve exact numbers. Never round, estimate, or "helpfully" infer a stitch count, coordinate, or dimension that isn't explicitly stated.
  • Strip, don't summarize. Isolation tasks remove noise; they don't paraphrase the signal.
  • Default to tables for anything comparative (boundaries, audits, gap-finding) — two or three columns max, mobile-width friendly.
  • Never invent CSS/spec values. If a dimension isn't in the source text, leave the field blank or flag it — don't fill gaps with typical defaults.
  • Don't leave marketing buzzwords partially sanitized — if "cloud-based" is removed but "always-on" survives, the pass is incomplete; re-scan.
  • Don't output a chronological task list when the workflow calls for the Anti-Fog single-task mode — that reintroduces option fatigue.
  • Don't "clean up" pattern text by correcting what looks like an error — flag it via Audit Row Invariants instead, let the human decide.
  • Don't merge the Isolation step and the Formatting step — isolate first (pure extraction), then format (presentation), so nothing is lost in translation.
0
Grade B+AI Skill Framework
Scorecard
Criteria Breakdown
Quick Start
13/15
Workflow
14/15
Examples
16/20
Completeness
16/20
Format
14/15
Conciseness
12/15