feat(bird_x): add noise words + last-chance retry with strongest token
Cherry-picked from PR #24 (el-analista). Adds trending/viral/plugin/skill/tool noise words to _extract_core_subject, and a last-chance retry that falls back to the longest non-noise token when 2-word retry also returns 0 results. cache.py and render.py env overrides were already on main. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
@@ -89,9 +89,11 @@ def _extract_core_subject(topic: str) -> str:
|
|||||||
# Research/meta descriptors
|
# Research/meta descriptors
|
||||||
'best', 'top', 'good', 'great', 'awesome', 'killer',
|
'best', 'top', 'good', 'great', 'awesome', 'killer',
|
||||||
'latest', 'new', 'news', 'update', 'updates',
|
'latest', 'new', 'news', 'update', 'updates',
|
||||||
|
'trendiest', 'trending', 'hottest', 'hot', 'popular', 'viral',
|
||||||
'practices', 'features', 'guide', 'tutorial',
|
'practices', 'features', 'guide', 'tutorial',
|
||||||
'recommendations', 'advice', 'review', 'reviews',
|
'recommendations', 'advice', 'review', 'reviews',
|
||||||
'usecases', 'examples', 'comparison', 'versus', 'vs',
|
'usecases', 'examples', 'comparison', 'versus', 'vs',
|
||||||
|
'plugin', 'plugins', 'skill', 'skills', 'tool', 'tools',
|
||||||
# Prompting meta words
|
# Prompting meta words
|
||||||
'prompt', 'prompts', 'prompting', 'techniques', 'tips',
|
'prompt', 'prompts', 'prompting', 'techniques', 'tips',
|
||||||
'tricks', 'methods', 'strategies', 'approaches',
|
'tricks', 'methods', 'strategies', 'approaches',
|
||||||
@@ -286,6 +288,21 @@ def search_x(
|
|||||||
_log(f"0 results for '{core_topic}', retrying with '{shorter}'")
|
_log(f"0 results for '{core_topic}', retrying with '{shorter}'")
|
||||||
query = f"{shorter} since:{from_date}"
|
query = f"{shorter} since:{from_date}"
|
||||||
response = _run_bird_search(query, count, timeout)
|
response = _run_bird_search(query, count, timeout)
|
||||||
|
items = parse_bird_response(response)
|
||||||
|
|
||||||
|
# Last-chance retry: use strongest remaining token (often the product name)
|
||||||
|
if not items and core_words:
|
||||||
|
low_signal = {
|
||||||
|
'trendiest', 'trending', 'hottest', 'hot', 'popular', 'viral',
|
||||||
|
'best', 'top', 'latest', 'new', 'plugin', 'plugins',
|
||||||
|
'skill', 'skills', 'tool', 'tools',
|
||||||
|
}
|
||||||
|
candidates = [w for w in core_words if w not in low_signal]
|
||||||
|
if candidates:
|
||||||
|
strongest = max(candidates, key=len)
|
||||||
|
_log(f"0 results for '{core_topic}', retrying with strongest token '{strongest}'")
|
||||||
|
query = f"{strongest} since:{from_date}"
|
||||||
|
response = _run_bird_search(query, count, timeout)
|
||||||
|
|
||||||
return response
|
return response
|
||||||
|
|
||||||
|
|||||||
@@ -0,0 +1,57 @@
|
|||||||
|
"""Tests for bird_x module."""
|
||||||
|
|
||||||
|
import sys
|
||||||
|
import unittest
|
||||||
|
from pathlib import Path
|
||||||
|
from unittest import mock
|
||||||
|
|
||||||
|
sys.path.insert(0, str(Path(__file__).parent.parent / "scripts"))
|
||||||
|
|
||||||
|
from lib import bird_x
|
||||||
|
|
||||||
|
|
||||||
|
class TestExtractCoreSubject(unittest.TestCase):
|
||||||
|
def test_strips_trending_noise(self):
|
||||||
|
result = bird_x._extract_core_subject("trendiest Claude Code skills")
|
||||||
|
self.assertNotIn("trendiest", result)
|
||||||
|
self.assertIn("claude", result.lower())
|
||||||
|
|
||||||
|
def test_strips_tool_noise(self):
|
||||||
|
result = bird_x._extract_core_subject("best AI tools for coding")
|
||||||
|
self.assertNotIn("tools", result)
|
||||||
|
self.assertNotIn("best", result)
|
||||||
|
|
||||||
|
def test_strips_skill_noise(self):
|
||||||
|
result = bird_x._extract_core_subject("top claude code skills")
|
||||||
|
self.assertNotIn("skills", result)
|
||||||
|
self.assertNotIn("top", result)
|
||||||
|
|
||||||
|
|
||||||
|
class TestBirdSearchRetries(unittest.TestCase):
|
||||||
|
def test_last_chance_retry_uses_strongest_token(self):
|
||||||
|
"""When shorter retry also returns 0, uses longest non-noise token."""
|
||||||
|
empty = {"items": []}
|
||||||
|
with mock.patch.object(bird_x, "_extract_core_subject", return_value="best codex skill plugin"), \
|
||||||
|
mock.patch.object(bird_x, "parse_bird_response", return_value=[]), \
|
||||||
|
mock.patch.object(bird_x, "_run_bird_search", return_value=empty) as run_mock:
|
||||||
|
bird_x.search_x("best codex skill plugin", "2026-01-01", "2026-01-31", depth="quick")
|
||||||
|
|
||||||
|
# Should try: original, shorter (2-word), last-chance (strongest token)
|
||||||
|
self.assertEqual(run_mock.call_count, 3)
|
||||||
|
queries = [call.args[0] for call in run_mock.call_args_list]
|
||||||
|
# Last call should use "codex" (longest non-noise word)
|
||||||
|
self.assertIn("codex", queries[2])
|
||||||
|
|
||||||
|
def test_no_retry_when_first_query_has_results(self):
|
||||||
|
"""No retry when first query succeeds."""
|
||||||
|
result = {"items": [{"id": "1"}]}
|
||||||
|
with mock.patch.object(bird_x, "_extract_core_subject", return_value="nano banana"), \
|
||||||
|
mock.patch.object(bird_x, "parse_bird_response", return_value=[{"id": "1"}]), \
|
||||||
|
mock.patch.object(bird_x, "_run_bird_search", return_value=result) as run_mock:
|
||||||
|
bird_x.search_x("nano banana prompting", "2026-01-01", "2026-01-31")
|
||||||
|
|
||||||
|
self.assertEqual(run_mock.call_count, 1)
|
||||||
|
|
||||||
|
|
||||||
|
if __name__ == "__main__":
|
||||||
|
unittest.main()
|
||||||
Reference in New Issue
Block a user