brand: meet Pip, the courier bird (#132)

* docs: add mascot & brand design spec (Pip the courier bird)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* docs: Pip wears a tie - update mascot spec to v5 flight loop

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* docs: add Pip brand PR implementation plan

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* docs(plan): v7 master GIF - drop frame scaling, add enclosed-hole transparency

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* docs(plan): v8 master GIF - fix hole classification (chest stays opaque)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* docs(plan): scrub stale v5 references

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* docs(plan): v10 master GIF - line-fitted envelope border clipping

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* docs(plan): label pipeline as v10

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* docs(plan): v16 master GIF - targeted removal of gap blob

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* docs(plan): v17 master GIF - drop envelope border clipping, keep blob removal

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* docs(plan): v19 final master GIF - user-approved thin outline repair

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat(brand): add Pip mascot assets and regeneration pipeline

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat(brand): Pip takes over the README header

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* docs(spec): scrub stale scaling line

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat(brand): avatar, social card, and mascot sources (PNG allowlist)

The global *.png personal-data rule silently excluded the mascot's source
sheets and generated PNGs; allowlist the upstream-controlled assets/mascot/
paths without weakening the fork-protecting rule.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Mads Lorentzen
2026-07-12 10:21:36 +02:00
committed by GitHub
co-authored by Claude Fable 5
parent 8bee3ddb97
commit 09f0417d78
17 changed files with 1042 additions and 1 deletions
+17
View File
@@ -0,0 +1,17 @@
# Pip, the courier bird
Brand assets for the project mascot. Design spec:
[`docs/superpowers/specs/2026-07-11-mascot-brand-design.md`](../../docs/superpowers/specs/2026-07-11-mascot-brand-design.md).
| File | Purpose |
|---|---|
| `pip_flight_loop.gif` | The animated brand mark (README header). 6 frames, transparent, <50 KB. |
| `pip_avatar.png` | Static square avatar (repo/social use). |
| `social_preview.png` | 1280x640 card for GitHub Settings -> Social preview. |
| `sources/` | AI-generated sprite sheets the frames were cut from. |
| `reference/` | Retired flat-style sprites, kept as palette reference. |
| `assemble_flight_loop.py` | Regenerates the GIF from the tie sheet (pillow+numpy+scipy). |
| `make_brand_assets.py` | Regenerates avatar + social card from the GIF (pillow). |
Palette (exactly 7 colors): `#2a9d8f` `#1f6f65` `#e76f51` `#969ba0` `#ffffff` `#e63946` `#22333b`.
The sprite never carries text or logos. Alt text is always "Pip, the courier bird".
+200
View File
@@ -0,0 +1,200 @@
"""Regenerate pip_flight_loop.gif from sources/chatgpt_tie_sheet.png.
Pipeline: 2D-cluster the sheet into 6 birds, snap every pixel to the 7-color
brand palette, order frames by wing centroid (no per-frame rescaling),
anchor on the beak, flood-fill edge-connected background to transparent.
Requires: pillow, numpy, scipy (maintainer tooling; not needed by CI).
"""
from pathlib import Path
from PIL import Image
import numpy as np
from scipy import ndimage
HERE = Path(__file__).parent
SHEET = HERE / "sources" / "chatgpt_tie_sheet.png"
OUT_GIF = HERE / "pip_flight_loop.gif"
PAL = np.array([
(42, 157, 143), (31, 111, 101), (231, 111, 81), (150, 155, 160),
(255, 255, 255), (230, 57, 70), (34, 51, 59),
], dtype=int)
TRANSPARENT = 7
CANVAS = 560
im = Image.open(SHEET).convert("RGB")
a_full = np.asarray(im).astype(int)
nonwhite_full = (a_full < 245).any(axis=2)
# 2D clustering: dilate to glue beak tips, label, keep 6 largest, sort by x
dil = ndimage.binary_dilation(nonwhite_full, iterations=2)
lab2, n2 = ndimage.label(dil)
sizes = ndimage.sum(dil, lab2, range(1, n2 + 1))
keep = np.argsort(sizes)[::-1][:6] + 1
comps = []
for k in keep:
mask = (lab2 == k) & nonwhite_full
ys, xs = np.where(mask)
comps.append(dict(mask=mask, x0=xs.min(), x1=xs.max(), y0=ys.min(), y1=ys.max(), cx=xs.mean()))
comps.sort(key=lambda d: d["cx"])
assert len(comps) == 6, f"expected 6 birds, found {len(comps)}"
def snap(arr):
d = ((arr[:, :, None, :] - PAL[None, None, :, :]) ** 2).sum(-1)
return PAL[d.argmin(-1)].astype(np.uint8)
cells = []
for comp in comps:
pad = 8
y0, y1 = max(0, comp["y0"] - pad), comp["y1"] + pad
x0, x1 = max(0, comp["x0"] - pad), comp["x1"] + pad
sub = a_full[y0:y1, x0:x1].copy()
submask = comp["mask"][y0:y1, x0:x1]
sub[~submask] = (255, 255, 255)
cells.append(snap(sub))
def color_mask(arr, ci, tol=10):
return np.abs(arr.astype(int) - PAL[ci]).sum(-1) < tol
infos = []
for i, c in enumerate(cells):
coral = color_mask(c, 2)
lab, n = ndimage.label(coral)
best, besty = None, 10 ** 9
for k in range(1, n + 1):
ys, xs = np.where(lab == k)
if len(ys) < 80:
continue
if ys.mean() < besty:
besty, best = ys.mean(), (xs, ys)
bx, by = best[0].mean(), best[1].mean()
dteal = color_mask(c, 1)
wing_c = float(np.where(dteal)[0].mean()) if dteal.sum() else by
infos.append(dict(i=i, beak=(bx, by), area=len(best[0]), wing_c=wing_c))
for d_ in infos:
# birds in the sheet are size-consistent (within 5%); rescaling by noisy
# beak measurements caused a visible zoom pulse in an earlier build
d_["scale"] = 1.0
d_["wing_rel"] = d_["wing_c"] - d_["beak"][1]
order_sorted = sorted(infos, key=lambda d: d["wing_rel"])
seq = [order_sorted[0], order_sorted[2], order_sorted[4], order_sorted[5], order_sorted[3], order_sorted[1]]
p_frames = []
for d_ in seq:
c = cells[d_["i"]]
h, w = c.shape[:2]
sc = d_["scale"]
c2 = np.asarray(Image.fromarray(c).resize((int(w * sc), int(h * sc)), Image.NEAREST))
beak = (d_["beak"][0] * sc, d_["beak"][1] * sc)
canvas = np.full((CANVAS, CANVAS, 3), 255, dtype=np.uint8)
tx, ty = int(CANVAS * 0.72), int(CANVAS * 0.38)
px, py = int(tx - beak[0]), int(ty - beak[1])
H2, W2 = c2.shape[:2]
x0, y0 = max(0, px), max(0, py)
x1, y1 = min(CANVAS, px + W2), min(CANVAS, py + H2)
canvas[y0:y1, x0:x1] = c2[y0 - py:y0 - py + (y1 - y0), x0 - px:x0 - px + (x1 - x0)]
nonwhite = (canvas < 250).any(axis=2)
lab, n = ndimage.label(nonwhite)
for k in range(1, n + 1):
ys, xs = np.where(lab == k)
cols = canvas[ys, xs].astype(int)
is_gray = (np.abs(cols - PAL[3]).sum(-1) < 60).mean() > 0.6
if len(ys) < 400 and ys.mean() < 300 and is_gray:
canvas[ys, xs] = (255, 255, 255)
nonwhite[ys, xs] = False
d2 = ((canvas.astype(int)[:, :, None, :] - PAL[None, None, :, :]) ** 2).sum(-1)
idx = d2.argmin(-1).astype(np.uint8)
white = ~nonwhite
wlab, wn = ndimage.label(white)
border = set(wlab[0, :]) | set(wlab[-1, :]) | set(wlab[:, 0]) | set(wlab[:, -1])
border.discard(0)
idx[np.isin(wlab, list(border))] = TRANSPARENT
# enclosed white pockets: classify by boundary composition. The chest patch
# always borders light teal (body interior); the envelope face is >=70%
# gray/red; true background pockets (between legs, body-envelope gap) are
# neither. NOTE: do not use outline/dark-teal ratios here - anti-aliased
# white-to-outline edges quantize to gray and poison those ratios.
for wl in range(1, wn + 1):
if wl in border:
continue
comp = wlab == wl
ring = ndimage.binary_dilation(comp, iterations=2) & ~comp & nonwhite
ridx = idx[ring]
if len(ridx) == 0:
continue
teal = (ridx == 0).sum() / len(ridx) # light teal only
envelope = np.isin(ridx, [3, 5]).sum() / len(ridx) # gray, red
if teal >= 0.08:
continue # chest: real content
if envelope >= 0.70:
continue # envelope face: real content, keep whole. (A border-clip
# was tried here to split merged face+gap components; it misfit
# tilted envelopes and bit into the face. The merged white reads
# fine as-is - do not reintroduce clipping.)
idx[comp] = TRANSPARENT
# targeted art cleanup: the source sheet has one large outlined teal blob
# (a vestigial appendage) drawn into the body-envelope gap of one frame.
# Only fragments >=100px qualify - smaller teal fragments are legitimate
# pixel-art texture, and generic cleanup rules damage the tie and feet.
bird_mask = np.isin(idx, [0, 1])
blab, bn = ndimage.label(bird_mask)
if bn > 1:
bsizes = ndimage.sum(bird_mask, blab, range(1, bn + 1))
bmain = int(np.argmax(bsizes)) + 1
gray_mask = idx == 3
glab, gn = ndimage.label(gray_mask)
a_fit, b_fit = 0.0, 10 ** 6
if gn:
gsizes = ndimage.sum(gray_mask, glab, range(1, gn + 1))
env_gray = glab == (int(np.argmax(gsizes)) + 1)
cols = np.where(env_gray.any(axis=0))[0]
tops = env_gray.argmax(axis=0)[cols].astype(float)
med = np.median(tops)
good = np.abs(tops - med) <= 20
if good.sum() >= 10:
a_fit, b_fit = np.polyfit(cols[good], tops[good], 1)
else:
a_fit, b_fit = 0.0, med
for bk in range(1, bn + 1):
if bk == bmain or not (100 <= bsizes[bk - 1] < 600):
continue
ys3, xs3 = np.where(blab == bk)
cy, cx = ys3.mean(), xs3.mean()
border_here = a_fit * cx + b_fit
if not (border_here - 55 <= cy < border_here):
continue
fmask = blab == bk
ring = ndimage.binary_dilation(fmask, iterations=2) & ~fmask
if not ring.any() or (idx[ring] == 6).mean() < 0.40:
continue
# erase the blob's own outline ring, but protect outline pixels
# that belong to the body, chest, tie, or feet (anything near the
# main structures) - an unprotected shell bit the body outline
protected = ndimage.binary_dilation(
(blab == bmain) | (idx == 4) | (idx == 2), iterations=2)
shell = ndimage.binary_dilation(fmask, iterations=3) & (idx == 6) & ~protected
erased = fmask | shell
idx[fmask] = TRANSPARENT
idx[shell] = TRANSPARENT
# repair: the blob's boundary doubled as the local body outline, so
# erasing it exposes bare chest white to the background. Repaint the
# exposed content edge (white/teal within the wound area, near the
# new transparency) as outline so the bird's underside stays closed.
near_wound = ndimage.binary_dilation(erased, iterations=6)
transp = idx == TRANSPARENT
near_gap = ndimage.binary_dilation(transp, iterations=6)
repaint = near_wound & near_gap & np.isin(idx, [0, 1, 4])
idx[repaint] = 6
p = Image.fromarray(idx, mode="P")
palette = PAL.astype(np.uint8).flatten().tolist() + [255, 0, 255]
p.putpalette(palette + [0] * (768 - len(palette)))
p.info["transparency"] = TRANSPARENT
p_frames.append(p)
p_frames[0].save(OUT_GIF, save_all=True, append_images=p_frames[1:],
duration=110, loop=0, disposal=2, transparency=TRANSPARENT, optimize=False)
print(f"wrote {OUT_GIF} ({OUT_GIF.stat().st_size} bytes)")
assert OUT_GIF.stat().st_size < 50_000, "README asset must stay under 50 KB"
+81
View File
@@ -0,0 +1,81 @@
"""Generate pip_avatar.png (512x512) and social_preview.png (1280x640)
from pip_flight_loop.gif. Requires pillow (maintainer tooling, not CI).
Layout rule for the card (the old draft's bug was text/mascot overlap):
mascot occupies x < 500; text starts at x = 540; an assertion enforces
that the widest text line fits inside the canvas.
"""
from pathlib import Path
from PIL import Image, ImageDraw, ImageFont
HERE = Path(__file__).parent
GIF = HERE / "pip_flight_loop.gif"
INK = (34, 51, 59) # 22333b
GRAY = (120, 126, 133)
SEAL = (230, 57, 70) # e63946
WORDMARK = "ai-job-search"
TAGLINE = "job search that runs on your machine"
def best_frame(gif_path, frame_index=0):
g = Image.open(gif_path)
g.seek(frame_index)
return g.convert("RGBA")
def find_font(size):
candidates = [
r"C:\Windows\Fonts\consolab.ttf", # Consolas Bold
r"C:\Windows\Fonts\consola.ttf",
"/usr/share/fonts/truetype/dejavu/DejaVuSansMono-Bold.ttf",
]
for c in candidates:
if Path(c).exists():
return ImageFont.truetype(c, size)
return ImageFont.load_default()
def make_avatar():
frame = best_frame(GIF, 0) # first frame: wing raised, reads best square
bbox = frame.getbbox()
crop = frame.crop(bbox)
side = max(crop.size) + 60
canvas = Image.new("RGBA", (side, side), (0, 0, 0, 0))
canvas.paste(crop, ((side - crop.width) // 2, (side - crop.height) // 2), crop)
canvas = canvas.resize((512, 512), Image.NEAREST)
out = HERE / "pip_avatar.png"
canvas.save(out)
print(f"wrote {out}")
def make_card():
W, H = 1280, 640
card = Image.new("RGB", (W, H), (255, 255, 255))
draw = ImageDraw.Draw(card)
frame = best_frame(GIF, 0)
bbox = frame.getbbox()
crop = frame.crop(bbox)
target_h = 420
scale = target_h / crop.height
crop = crop.resize((int(crop.width * scale), target_h), Image.NEAREST)
mascot_x = 90
assert mascot_x + crop.width <= 500, "mascot must stay left of x=500"
card.paste(crop, (mascot_x, (H - crop.height) // 2), crop)
f_big = find_font(72)
f_small = find_font(32)
tx = 540
draw.text((tx, 240), WORDMARK, font=f_big, fill=INK)
wm_w = draw.textlength(WORDMARK, font=f_big)
tag_w = draw.textlength(TAGLINE, font=f_small)
assert tx + max(wm_w, tag_w) <= W - 40, "text overflows card"
draw.text((tx, 340), TAGLINE, font=f_small, fill=GRAY)
# seal-red accent: small dot echoing the envelope seal
draw.ellipse((tx + 2, 402, tx + 22, 422), fill=SEAL)
draw.text((tx + 34, 400), "free & open source", font=f_small, fill=GRAY)
out = HERE / "social_preview.png"
card.save(out)
print(f"wrote {out}")
if __name__ == "__main__":
make_avatar()
make_card()
Binary file not shown.

After

Width:  |  Height:  |  Size: 8.1 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 29 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 2.4 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 2.6 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 1.6 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 28 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 882 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 762 KiB