Spaces:
Sleeping
Sleeping
File size: 10,170 Bytes
5ecf925 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 | """
Verify insertions tab β step through CR/TS screenshot pairs from a review
package (produced locally by the `cr-visual-pair` skill) and record OK /
Not OK / Wrong match verdicts. Verdicts live in st.session_state only; the
user downloads a text log at the end.
"""
import base64
import io
import json
import mimetypes
import zipfile
from pathlib import Path
import streamlit as st
VERDICT_LABELS = {
"ok": "OK",
"not_ok": "NOT OK",
"wrong_match": "WRONG MATCH (page shown did not correspond)",
}
SKILL_PATH = Path(__file__).parent / "skills" / "cr-visual-pair" / "SKILL.md"
_ZOOM_CSS = """
<style>
.cr-zoom-thumb { cursor: zoom-in; display: block; }
.cr-zoom-thumb img { max-width: 100%; border-radius: 4px; display: block; }
.cr-zoom-overlay {
display: none;
position: fixed;
inset: 0;
background: rgba(0, 0, 0, 0.9);
z-index: 999999;
align-items: center;
justify-content: center;
}
.cr-zoom-overlay:target { display: flex; }
.cr-zoom-overlay img {
max-width: 95vw;
max-height: 95vh;
object-fit: contain;
cursor: zoom-out;
}
.cr-zoom-close {
position: fixed;
top: 16px;
right: 24px;
color: #fff;
font-size: 2.5rem;
text-decoration: none;
line-height: 1;
}
</style>
"""
def _img_data_uri(path: Path) -> str:
mime = mimetypes.guess_type(path.name)[0] or "image/png"
data = base64.b64encode(path.read_bytes()).decode()
return f"data:{mime};base64,{data}"
def _render_zoomable_image(img_path: Path, anchor_id: str) -> None:
anchor_id = "".join(c if c.isalnum() else "-" for c in anchor_id)
data_uri = _img_data_uri(img_path)
st.markdown(
f"""
<a class="cr-zoom-thumb" href="#{anchor_id}">
<img src="{data_uri}">
</a>
<div class="cr-zoom-overlay" id="{anchor_id}">
<a class="cr-zoom-close" href="#">×</a>
<a href="#"><img src="{data_uri}"></a>
</div>
""",
unsafe_allow_html=True,
)
def _render_skill_card() -> None:
with st.expander("π’ cr-visual-pair"):
if not SKILL_PATH.exists():
st.error(f"Skill file not found at {SKILL_PATH}")
return
content = SKILL_PATH.read_text()
st.caption(
"Run this skill locally (Windows + Word) to produce a `review_package.zip` "
"from an ApplyCRs run. Save it as `~/.claude/skills/cr-visual-pair/SKILL.md`, "
"or download it below."
)
st.download_button(
"β¬ Download SKILL.md",
data=content,
file_name="SKILL.md",
mime="text/markdown",
key="verify_skill_download",
)
st.code(content, language="markdown")
def _extract_package(uploaded, dest_dir: Path) -> dict | None:
dest_dir.mkdir(parents=True, exist_ok=True)
zip_bytes = uploaded.getbuffer()
with zipfile.ZipFile(io.BytesIO(zip_bytes)) as zf:
zf.extractall(dest_dir)
manifest_path = dest_dir / "manifest.json"
if not manifest_path.exists():
return None
return json.loads(manifest_path.read_text())
def _build_log(manifest: dict, verdicts: dict, review_dir: Path) -> str:
lines = []
for pair in manifest.get("pairs", []):
pid = pair["id"]
verdict = verdicts.get(pid)
verdict_label = VERDICT_LABELS.get(verdict, "UNREVIEWED")
ts_spec = pair.get("ts_spec", "?")
ts_version = pair.get("ts_version", "?")
section = pair.get("section_number", "?")
match_note = "" if pair.get("matched", True) else " [auto-flagged: low-confidence match]"
lines.append(
f"TS {ts_spec} v{ts_version} [section {section}] : {verdict_label}{match_note}"
)
for change in pair.get("changes", []):
change_note = "" if change.get("matched", True) else " [low-confidence match]"
lines.append(
f" - CR {change.get('cr_uid', '?')} ({change.get('change_type', '?')})"
f"{change_note}"
)
return "\n".join(lines) + "\n"
def render_verify_tab(session_dir_fn, sid: str) -> None:
st.subheader("Verify insertions")
st.caption(
"Upload a review package (generated locally with the `cr-visual-pair` skill) "
"and step through each CR β TS screenshot pair to confirm the insertion is correct."
)
_render_skill_card()
st.markdown(_ZOOM_CSS, unsafe_allow_html=True)
review_dir = session_dir_fn(sid) / "review"
manifest_key = "verify_manifest"
review_dir_key = "verify_review_dir"
if manifest_key not in st.session_state:
uploaded = st.file_uploader(
"Upload review package (.zip)", type=["zip"], key="verify_upload"
)
if uploaded is not None:
manifest = _extract_package(uploaded, review_dir)
if manifest is None:
st.error("This zip does not contain a manifest.json β not a valid review package.")
return
st.session_state[manifest_key] = manifest
st.session_state[review_dir_key] = str(review_dir)
st.session_state["verify_idx"] = 0
st.session_state["verify_verdicts"] = {}
st.rerun()
return
manifest = st.session_state[manifest_key]
pairs = manifest.get("pairs", [])
review_dir = Path(st.session_state[review_dir_key])
if not pairs:
st.warning("This review package has no pairs to review.")
if st.button("Upload a different package"):
for k in (manifest_key, review_dir_key, "verify_idx", "verify_verdicts"):
st.session_state.pop(k, None)
st.rerun()
return
idx = st.session_state.setdefault("verify_idx", 0)
verdicts = st.session_state.setdefault("verify_verdicts", {})
idx = max(0, min(idx, len(pairs) - 1))
st.session_state["verify_idx"] = idx
pair = pairs[idx]
pid = pair["id"]
n_ok = sum(1 for v in verdicts.values() if v == "ok")
n_not_ok = sum(1 for v in verdicts.values() if v == "not_ok")
n_wrong = sum(1 for v in verdicts.values() if v == "wrong_match")
n_reviewed = len(verdicts)
st.caption(
f"Pair {idx + 1} / {len(pairs)} Β· "
f"Reviewed: {n_reviewed}/{len(pairs)} Β· β
{n_ok} β {n_not_ok} β {n_wrong}"
)
st.markdown(
f"**TS** `{pair.get('ts_spec', '?')}` v{pair.get('ts_version', '?')} Β· "
f"section `{pair.get('section_number', '?')}`"
)
if not pair.get("matched", True):
st.warning(
"β The pairing tool could not confidently locate this section's page in the "
"TS β double-check whether it's the right page before judging the insertion."
)
existing_verdict = verdicts.get(pid)
if existing_verdict:
st.info(f"Current verdict: **{VERDICT_LABELS[existing_verdict]}**")
col_cr, col_ts = st.columns(2)
with col_cr:
st.markdown("**CR (source change)**")
changes = pair.get("changes", [])
if changes:
for change in changes:
label = f"`{change.get('cr_uid', '?')}` Β· {change.get('change_type', '?')}"
if not change.get("matched", True):
label += " β low-confidence"
st.caption(label)
if change.get("summary"):
st.caption(change["summary"])
cr_images = change.get("cr_images", [])
if cr_images:
for img_idx, img in enumerate(cr_images):
img_path = review_dir / img
if img_path.exists():
_render_zoomable_image(
img_path, f"cr-zoom-{pid}-{change.get('cr_uid', '?')}-{img_idx}"
)
else:
st.error(f"Missing image: {img}")
else:
st.warning("No CR image for this change.")
else:
st.warning("No CR changes recorded for this pair.")
with col_ts:
st.markdown("**TS (applied output)**")
ts_images = pair.get("ts_images", [])
if ts_images:
for img_idx, img in enumerate(ts_images):
img_path = review_dir / img
if img_path.exists():
_render_zoomable_image(img_path, f"ts-zoom-{pid}-{img_idx}")
else:
st.error(f"Missing image: {img}")
else:
st.warning("No TS image for this pair β pairing tool did not find a candidate page.")
def _advance():
if st.session_state["verify_idx"] < len(pairs) - 1:
st.session_state["verify_idx"] += 1
col1, col2, col3, col4, col5 = st.columns(5)
with col1:
if st.button("β
OK", type="primary", key=f"ok_{pid}"):
verdicts[pid] = "ok"
_advance()
st.rerun()
with col2:
if st.button("β Not OK", key=f"not_ok_{pid}"):
verdicts[pid] = "not_ok"
_advance()
st.rerun()
with col3:
if st.button("β Wrong match", key=f"wrong_{pid}"):
verdicts[pid] = "wrong_match"
_advance()
st.rerun()
with col4:
if st.button("β Prev", disabled=idx == 0, key=f"prev_{pid}"):
st.session_state["verify_idx"] -= 1
st.rerun()
with col5:
if st.button("Next β", disabled=idx == len(pairs) - 1, key=f"next_{pid}"):
st.session_state["verify_idx"] += 1
st.rerun()
if idx == len(pairs) - 1 and pid in verdicts:
st.success("Review complete β all pairs have a verdict. Download the log below.")
st.divider()
log_text = _build_log(manifest, verdicts, review_dir)
st.download_button(
"β¬ Download review log",
data=log_text,
file_name=f"cr_review_{sid[:8]}.log",
mime="text/plain",
)
if st.button("Upload a different package", key="verify_reset"):
for k in (manifest_key, review_dir_key, "verify_idx", "verify_verdicts"):
st.session_state.pop(k, None)
st.rerun()
|