fix: strip count header in reader_mode, trim README to sentence boundary, graceful reskill 403 (#139)
Fix 1a: In render_press_context(), when reader_mode=True, strip the stale 'N repos have press correlation:' count header via re.sub after template rendering. Narrative paragraphs are self-contained; the raw count line is AI-prompt noise that leaks into reader output. Fix 1b: _extract_readme_description() now trims each candidate line to the last complete sentence boundary (. ! ?) before returning it. Lines with no sentence boundary are skipped entirely rather than returning a truncated mid-sentence fragment. Removes the 150-char upper-bound check (sentence trimming makes it redundant). Fix 2: reskill.py main() now wraps call_github_models() in try/except RuntimeError. On 403 (no model access) or any other API failure, it writes a placeholder reskill report and exits 0 rather than crashing the job. Tests: 6 new tests added. 519 total pass. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Juan Manuel Servera committed
May 19, 2026 at 23:03 UTC
71e1c1cefd2e8c20cc8b997807f6f442f7566190
4 files changed
+144
-4
scripts/render_press_context.py
+20
-2
@@ -78,6 +78,9 @@ def _extract_readme_description(snippet: str) -> str:
78
"""Return the first readable descriptive line from a README snippet.
79
80
Skips headings, badge lines, image tags, and blank lines.
81
+ Trims each candidate line to the last complete sentence boundary
82
+ (. ! ?) so truncated snippets never show a broken mid-sentence tail.
83
+ If no sentence boundary is found within a line, that line is skipped.
84
Returns an empty string if nothing usable is found.
85
"""
86
for line in snippet.splitlines():
@@ -91,8 +94,20 @@ def _extract_readme_description(snippet: str) -> str:
94
line = re.sub(r"\[([^\]]+)\]\([^)]+\)", r"\1", line)
95
line = re.sub(r"<[^>]+>", "", line)
96
line = re.sub(r"[*_`>]", "", line)
94
- line = line.strip(" .,;:")
95
- if 20 <= len(line) <= 150:
97
+ line = line.strip()
98
+ if not line:
99
+ continue
100
+ # Trim to the last complete sentence boundary (. ! ? followed by space or end)
101
+ last_boundary = -1
102
+ for i in range(len(line) - 1, -1, -1):
103
+ if line[i] in ".!?" and (i + 1 >= len(line) or line[i + 1] == " "):
104
+ last_boundary = i
105
+ break
106
+ if last_boundary < 0:
107
+ # No sentence boundary — drop this line to avoid broken sentences
108
+ continue
109
+ line = line[: last_boundary + 1].strip(" .,;:")
110
+ if len(line) >= 20:
111
return line
112
return ""
113
@@ -546,6 +561,9 @@ def render_press_context(
561
instructions_marker = "\n### Instructions\n"
562
if instructions_marker in rendered:
563
rendered = rendered[: rendered.index(instructions_marker)]
564
+ # Remove the count header "N repos have press correlation:" — narrative
565
+ # paragraphs are self-contained; the raw count line is noise in reader mode.
566
+ rendered = re.sub(r"\d+ repos have press correlation:\n", "", rendered)
567
568
# Append divergences section
569
divergence_section = format_divergences(divergences, reader_mode=reader_mode)
scripts/reskill.py
+12
-2
@@ -24,7 +24,7 @@ DEFAULT_WISDOM_FILE = ROOT / ".squad" / "identity" / "wisdom.md"
24
DEFAULT_SKILLS_DIR = ROOT / ".squad" / "skills"
25
DEFAULT_REPORT_DIR = ROOT / ".squad" / "reskill"
26
DEFAULT_MODELS_ENDPOINT = "https://models.github.ai/inference/chat/completions"
27
-DEFAULT_MODELS_MODEL = "openai/gpt-4.1"
27
+DEFAULT_MODELS_MODEL = "openai/gpt-4o"
28
DEFAULT_MODELS_TIMEOUT = 30
29
30
@@ -300,7 +300,17 @@ def main(argv: list[str] | None = None) -> int:
300
args.prompt_output.parent.mkdir(parents=True, exist_ok=True)
301
args.prompt_output.write_text(prompt, encoding="utf-8")
302
303
- markdown = call_github_models(prompt)
303
+ try:
304
+ markdown = call_github_models(prompt)
305
+ except RuntimeError as exc:
306
+ print(f"⚠️ Reskill: GitHub Models call failed — {exc}", file=sys.stderr)
307
+ print("Writing placeholder reskill report and continuing.", file=sys.stderr)
308
+ output_path.parent.mkdir(parents=True, exist_ok=True)
309
+ output_path.write_text(
310
+ f"# Reskill skipped\n\nGitHub Models unavailable: {exc}\n",
311
+ encoding="utf-8",
312
+ )
313
+ return 0
314
output_path.parent.mkdir(parents=True, exist_ok=True)
315
output_path.write_text(markdown, encoding="utf-8")
316
return 0
tests/test_render_press_context.py
+52
@@ -516,3 +516,55 @@ class TestFormatCorrelationsNarrative:
516
result = render_press_context(tc, corr_data, "2026-W21", reader_mode=False)
517
assert "confidence:" in result
518
assert "### Instructions" in result
519
+
520
+
521
+class TestReaderModeCountHeader:
522
+ """reader_mode=True must not emit the raw 'N repos have press correlation:' header."""
523
+
524
+ def _corrs(self, n: int = 5) -> list[dict]:
525
+ return [
526
+ {
527
+ "repo": f"openai/repo-{i}",
528
+ "match_type": "org_name",
529
+ "correlation_confidence": 0.8,
530
+ "hype_risk": "low",
531
+ "matched_articles": [],
532
+ }
533
+ for i in range(n)
534
+ ]
535
+
536
+ def test_reader_mode_omits_count_header(self):
537
+ tc = {"articles": []}
538
+ corr_data = {"correlations": self._corrs(5), "divergences": {}}
539
+ result = render_press_context(tc, corr_data, "2026-W21", reader_mode=True)
540
+ assert "repos have press correlation" not in result
541
+
542
+ def test_ai_mode_keeps_count_header(self):
543
+ tc = {"articles": []}
544
+ corr_data = {"correlations": self._corrs(5), "divergences": {}}
545
+ result = render_press_context(tc, corr_data, "2026-W21", reader_mode=False)
546
+ assert "repos have press correlation" in result
547
+
548
+
549
+class TestExtractReadmeDescriptionSentenceBoundary:
550
+ """_extract_readme_description must not return mid-sentence truncated text."""
551
+
552
+ def test_drops_line_without_sentence_boundary(self):
553
+ # Simulates a 500-char truncation mid-sentence
554
+ snippet = "# Guava\n\nGuava is a set of core Java libraries from Google that includes new collect"
555
+ result = _extract_readme_description(snippet)
556
+ assert result == ""
557
+
558
+ def test_trims_to_last_sentence_in_long_line(self):
559
+ snippet = (
560
+ "# Lib\n\n"
561
+ "This library does X. It also does Y. And even more beyond that without end"
562
+ )
563
+ result = _extract_readme_description(snippet)
564
+ # Should trim to the last complete sentence boundary
565
+ assert result == "This library does X. It also does Y"
566
+
567
+ def test_returns_empty_when_no_boundary_in_snippet(self):
568
+ snippet = "# Header\n\nNo period here at all and the line is long enough to match normally"
569
+ result = _extract_readme_description(snippet)
570
+ assert result == ""
tests/test_reskill.py
+60
@@ -170,6 +170,66 @@ class ReskillTests(unittest.TestCase):
170
self.assertTrue(prompt_output_path.exists())
171
self.assertIn("Prefer durable signals.", prompt_output_path.read_text(encoding="utf-8"))
172
173
+ def test_main_handles_model_403_gracefully(self) -> None:
174
+ """When GitHub Models returns 403 (no model access), main exits 0 with a placeholder."""
175
+ tests_root = Path(__file__).resolve().parent
176
+ with tempfile.TemporaryDirectory(dir=tests_root) as tmpdir:
177
+ base = Path(tmpdir)
178
+ analyzed_dir = base / "data" / "analyzed"
179
+ snapshots_dir = base / "data" / "snapshots"
180
+ wisdom_path = base / ".squad" / "identity" / "wisdom.md"
181
+ skills_dir = base / ".squad" / "skills"
182
+ prompt_template = base / "reskill.md"
183
+ output_path = base / ".squad" / "reskill" / "2026-W21.md"
184
+ analyzed_dir.mkdir(parents=True)
185
+ snapshots_dir.mkdir(parents=True)
186
+ wisdom_path.parent.mkdir(parents=True)
187
+ skills_dir.mkdir(parents=True)
188
+ output_path.parent.mkdir(parents=True)
189
+ wisdom_path.write_text("# Wisdom\n\nPrefer durable signals.", encoding="utf-8")
190
+ prompt_template.write_text("{{WISDOM}}\n{{QUALITY_TREND}}", encoding="utf-8")
191
+
192
+ from urllib import error as urlerror
193
+ import io as _io
194
+
195
+ fake_body = _io.BytesIO(
196
+ b'{"error":{"code":"no_access","message":"No access to model: openai/gpt-4.1"}}'
197
+ )
198
+ http_err = urlerror.HTTPError(
199
+ url="https://models.github.ai",
200
+ code=403,
201
+ msg="Forbidden",
202
+ hdrs={}, # type: ignore[arg-type]
203
+ fp=fake_body,
204
+ )
205
+
206
+ with mock.patch.dict("os.environ", {"GITHUB_TOKEN": "token"}, clear=False), mock.patch.object(
207
+ reskill.request, "urlopen", side_effect=http_err
208
+ ):
209
+ exit_code = reskill.main(
210
+ [
211
+ "--current-datetime",
212
+ "2026-05-18T15:22:25.067+02:00",
213
+ "--prompt-template",
214
+ str(prompt_template),
215
+ "--analyzed-dir",
216
+ str(analyzed_dir),
217
+ "--snapshots-dir",
218
+ str(snapshots_dir),
219
+ "--wisdom-file",
220
+ str(wisdom_path),
221
+ "--skills-dir",
222
+ str(skills_dir),
223
+ "--output",
224
+ str(output_path),
225
+ ]
226
+ )
227
+
228
+ self.assertEqual(exit_code, 0)
229
+ content = output_path.read_text(encoding="utf-8")
230
+ self.assertIn("Reskill skipped", content)
231
+ self.assertIn("403", content)
232
+
233
234
if __name__ == "__main__":
235
unittest.main()