What does Logo Design · Full Brand Mark Pipeline do?
The core value of this skill is not 'being able to draw a logo,' but 'not repeating the same 21 mistakes.' It was distilled from 11 real-world iterations of the PH-002 Stampcrate project (each round was thrown out), turning failure modes, physical limits, and review questions into reference tables.
§0 Triage Router
| Symptom (what the user says) | Which Path | Must Read |
|---|---|---|
| "Make me a logo / design a brand mark" | Full pipeline (§2 Six Stages) | references/02-Metaphor_and_Category.md + references/03-Geometric_Constants.md |
| "I'm not happy with this logo, but I can't say why" | Diagnosis (§3) | references/01-Failure_Catalog.md (go through 21 items one by one) |
| "Several revisions already, each one more bland" | Change metaphor (§4) | references/06-Iteration_Protocol.md |
| "Logo is decided, help me produce full assets" | Delivery (§5) | references/04-Engineering_Compliance.md |
| "Let Claude/someone review this" | Review Pack (§6) | references/05-Review_Protocol.md |
Default judgment: When the user says "not satisfied," first determine if it's a 'direction error' or a 'craft error' — these two require completely opposite fixes. Getting them mixed up will waste five rounds.
§1 Three Non-Negotiable Hard Rules
1.1 Silhouette + Fine Detail = Small Size Disaster
This was the cause of death in 8 out of 11 rounds. One sentence summary:
Symbols where information is in the 'color-block shape' naturally survive at 16px; symbols where information is in 'silhouette + details' will inevitably break at 16px.
Bad example: Hand-held stamp silhouette (handle + neck + body + teeth) → teeth hit the frame, lip becomes two balls, teeth clearance gaps turn into "legs" Good example: Solid circle + white text → at 16px it reads as "red circle + white text," no degradation needed
First question before you start: Does this symbol's recognizability rely on the overall color block or on internal details? If the latter, prepare for five rounds.
1.2 All Details Scale by u = 100 / size
In viewBox="0 0 100 100", 1 display pixel = u viewBox units.
u = 100 / size
stroke width ≥ 3.0 * u # below this, disappears at 32px
clearance ≥ 2.5 * u # between inner frame and outer teeth
teeth amplitude ≥ 1.4 * u # below this, 16px effectively has no teeth
Most common mistake: Making only the "stroke" scale with size, but forgetting to scale the elements it squeezes. This is how Stampcrate v2 died — the inner frame was fixed at x=19, the stroke grew from 4.68 to 11.00, and the gap was compressed to 1.00 unit (= 0.16 pixels at 16px, completely overlapping).
1.3 Cuts Must Be Rounded, Strokes Must Be Thick Enough
- Sharp corners/serrations at 24px will be read as "legs," "spikes," or "crowns" — use rounded cuts or square cuts
- Stroke thinner than 3% (100×100) at 32px will disappear, looking "fuzzy, unprofessional"
These two rules come from external methodology (op7418/logo-generator-skill Part 0), validated by our tests.
§2 Six-Stage Pipeline
Stage 1 · Define Requirements (Ask 4 Questions, No More)
- What is the product? At what moment does the user see this mark (browser tab / header / packaging / business card)
- What is the core semantic? (One sentence, verb-first: connect, protect, finalize, collect)
- What cultural circle is the target market? (Determines if typographic symbols can be used — see
02-Metaphor_and_Category.md) - Are there explicitly excluded categories? (Determines which symbols are absolutely forbidden)
Stage 2 · Choose a Metaphor (Easiest to Skip, Most Decisive)
Don't draw "what the product is." First ask:
What outcome does the user actually want? Drawing the "outcome" is almost always stronger than drawing the "object."
Stampcrate's example: Selling stamps → draw a stamp (11 rounds stuck in that problem) → changed to drawing "a seal/emblem" — that broke through.
See references/02-Metaphor_and_Category.md (includes cultural misalignment decision table).
Stage 3 · Generate Variants (6+ at Once, Don't "Iterate One at a Time")
- At least 6 directions, each must be describable in one sentence: "This expresses X"
- Cover: concrete / abstract / letter / text / pure geometric / combination
- Render all at 128 / 32 / 16 three sizes side-by-side (this single image is your decision basis)
Use assets/render_pipeline.py, don't hand-craft.
Stage 4 · Physical Lower-Bound Validation (Mechanical, Not Aesthetic)
Run each candidate through:
☐ Can you still recognize it at 16px?
☐ Stroke ≥ 3u?
☐ Clearance ≥ 2.5u?
☐ Are cuts rounded?
☐ Negative space ≥ 40%?
☐ Elements ≤ 2? (If more, justify why the category allows it)
Look up directly in references/03-Geometric_Constants.md tables.
Stage 5 · Compliance (Skip this, and the printer will reject it)
☐ All text converted to paths (no <text> or font-family in SVG)
☐ Font licenses allow commercial use (prefer OFL; system fonts with unclear licenses are forbidden)
☐ Small-size exports are real exported PNGs, not browser scaling
☐ Favicon suite complete (ico 16/32/48 + svg + apple-touch 180 + 192/512 + webmanifest)
☐ Monochrome / reverse / dark-background versions all exist
See references/04-Engineering_Compliance.md.
Stage 6 · Review (The Question Matters More Than the Answer)
Don't ask "Is this good?" Ask "Which one is closest + what's still missing?" Review pack format in references/05-Review_Protocol.md.
§3 Diagnosis: When the User Says "Not Satisfied"
Go through the 21 items in references/01-Failure_Catalog.md one by one. First classify the type:
| Symptom | Type | Fix |
|---|---|---|
| "Looks like X" (pawn/pushpin/Pac-Man/hamburger menu) | Shape misreading | Adjust proportions, not direction |
| "Too small and becomes blurry / has legs / has spikes" | Physical limit | Thicken strokes, round corners, reduce elements |
| "Reads as another category" | Cultural/category misalignment | Change the symbol, don't patch |
| "Can't say why, just not satisfied" | Missing character | See §4 |
| "Each round gets more bland" | Defensive iteration | See §4 |
§4 Change Metaphor: When Iteration Becomes "More Bland"
Signal: 3+ consecutive rounds of "patching over the last criticism," each fix stripping away some character.
This is a method error, not a direction error. Three moves to break out:
- Change the subject: Don't draw "what the product is," draw "what happens when the user gets it"
- Swap half the brand name: If the brand name is a compound word (e.g., stampcrate), check if half of it has never entered the visual
- Move character to the wordmark: Keep the mark neutral, let a custom wordmark carry the personality (many mature brands do this)
See references/06-Iteration_Protocol.md.
§5 Delivery: Asset Inventory
assets/logo/
logo-mark-l1.svg ≥64px full version
logo-mark-l2.svg 32–56px compact version
logo-mark-l3.svg ≤24px minimum version
logo-mark-reverse.svg reverse on dark background
logo-mark-mono.svg monochrome (print)
logo-lockup.svg horizontal lockup (light)
logo-lockup-dark.svg horizontal lockup (dark)
favicon/ 6-piece set
png/ real exported PNGs at each size
_fonts/OFL.txt font license copy
Litmus test: View favicon-16.png alone — if you can identify it, it passes.
§6 Resource Index (Load on Demand, Do Not Read All)
| File | Content | When to Read |
|---|---|---|
references/01-Failure_Catalog.md | 21 failure modes: symptom / root cause / fix / from which round | Required for diagnosis; read once before starting to save the most time |
references/02-Metaphor_and_Category.md | Metaphor selection method + cultural misalignment decision table + consistency principle | Required for Stage 2 |
references/03-Geometric_Constants.md | Physical limit table (stroke/clearance/teeth/whitespace/size tiers) + formulas | Required for Stage 4 |
references/04-Engineering_Compliance.md | Path conversion / font licensing / real exports / favicon suite / check scripts | Required for Stage 5 |
references/05-Review_Protocol.md | Review pack structure + 5 mandatory questions + 2 premise-challenging questions | Required for Stage 6 |
references/06-Iteration_Protocol.md | Defensive iteration identification + 3 breakthrough moves + when to stop | Read when stuck |
assets/render_pipeline.py | Reusable renderer skeleton (multi-size + real export + review sheet) | Use directly in Stage 3 |
examples/stampcrate-11-round-postmortem.md | Full real case: death cause and fix for each of the 11 rounds | Read for details |
CHANGELOG.md | Continuous iteration record: add experience after each new project | Update after each use |
§7 Continuous Iteration Protocol (How This Skill Grows)
After completing each new logo project, you must do three things (this is the only mechanism that makes the skill keep improving):
- Add a failure mode: If this project's mistake is not in the 21 items of document 01, add item 22 (symptom / root cause / fix / source project)
- Calibrate physical constants: If measured values in this project don't match document 03, update the table and note "XX project measurement"
- Update CHANGELOG: Write one line (date / project / what was added / what was overturned)
Forbidden: Writing records like "this round went smoothly" with no information value. Only record things that will 'prevent it next time.'
§8 Three Red Lines
- Do not skip Stage 2 (metaphor). Skipping it means all subsequent rounds work on the wrong problem.
- Do not substitute "looks better" for physical lower-bound validation. Aesthetics can be debated; "can you recognize it at 16px" cannot.
- Do not declare small-size usability without real exported PNGs. Browser scaling lies.