[eric] onboarding: sharpen the scan-driven reveal (surface telling apps, insight-led greeting)

backend/apps/onboarding/local_scan.py

@@ -21,6 +21,18 @@ SCAN_FOLDERS = ("Downloads", "Desktop", "Documents")
 REPO_PARENT_CANDIDATES = ("dev", "code", "projects", "src", "repos", "Documents/GitHub")
 SCREENSHOT_PREFIXES = ("screenshot", "screen shot", "screen recording")
 
+# The high-signal tools that actually tell prep who this person is (an IDE, a design app, a DAW) so the greeting can lead with a confident read instead of drowning in Calculator + System Settings. Matched by exact name or "name " prefix so "Arc" never trips on "Search".
+NOTABLE_APP_KEYWORDS = (
+    "xcode", "visual studio code", "cursor", "zed", "sublime text", "intellij idea",
+    "pycharm", "webstorm", "android studio", "docker", "orbstack", "postman", "iterm",
+    "warp", "ghostty", "tableplus", "github desktop", "figma", "sketch",
+    "adobe photoshop", "adobe illustrator", "adobe xd", "affinity", "framer", "blender",
+    "cinema 4d", "final cut pro", "davinci resolve", "adobe premiere pro",
+    "adobe after effects", "screenflow", "capcut", "obs", "logic pro", "ableton live",
+    "fl studio", "garageband", "unity", "unreal engine", "godot", "notion", "obsidian",
+    "ulysses", "scrivener", "craft", "linear", "tableau", "rstudio", "ollama", "lm studio", "arc",
+)
+
 
 @typechecked
 def p_list_apps() -> List[str]:
@@ -66,6 +78,16 @@ def p_summarize_folder(folder: Path) -> FolderSummary:
     return summary
 
 
+@typechecked
+def detect_signal_apps(apps: List[str]) -> List[str]:
+    out: List[str] = []
+    for name in apps:
+        low = name.lower()
+        if any(low == kw or low.startswith(kw + " ") for kw in NOTABLE_APP_KEYWORDS):
+            out.append(name)
+    return out[:20]
+
+
 @typechecked
 def p_count_git_repos(home: Path) -> int:
     count = 0
@@ -90,8 +112,10 @@ def p_count_git_repos(home: Path) -> int:
 @typechecked
 def run_local_scan(home: Path) -> ScanResult:
     folders = [p_summarize_folder(home / name) for name in SCAN_FOLDERS]
+    apps = p_list_apps()
     return ScanResult(
-        apps=p_list_apps(),
+        apps=apps,
+        signal_apps=detect_signal_apps(apps),
         folders=folders,
         git_repo_count=p_count_git_repos(home),
         has_gitconfig=(home / ".gitconfig").is_file(),

backend/apps/onboarding/models.py

@@ -35,6 +35,8 @@ class ScanResult(BaseModel):
     model_config = ConfigDict(validate_assignment=True)
 
     apps: List[str] = Field(default_factory=list)
+    # The high-signal subset of apps (IDEs, design/creative tools); the profile leans on these.
+    signal_apps: List[str] = Field(default_factory=list)
     folders: List[FolderSummary] = Field(default_factory=list)
     git_repo_count: int = 0
     has_gitconfig: bool = False

backend/apps/onboarding/prep.py

@@ -28,9 +28,11 @@ P_SYSTEM = (
     "organize local files, browse the web in a real browser, build small apps, and run agents in parallel. "
     "Given facts about the user's machine and the apps they picked, respond with STRICT JSON only: "
     '{"greeting": string, "starters": [{"title": string, "prompt": string, "reason": string}], "app_title": string, "app_prompt": string, "app_reason": string, "automations": [{"title": string, "prompt": string, "cadence": "daily"|"weekday"|"weekly"}]}. '
-    "First, silently infer a short profile of this user. If usage_summary is present it is the STRONGEST signal (it is "
-    "what they actually ask their AI about and facts their AI remembers about them); weight it above everything else, "
-    "then apps, folders, plan tier, email domain. Tune every task and the personal app to that profile; do not output the profile. "
+    "First, silently infer a short, confident profile of this user: who they are and what they are working on. "
+    "If usage_summary is present it is the STRONGEST signal (it is what they actually ask their AI about and facts "
+    "their AI remembers about them); weight it above everything else, then signal_apps (the high-signal tools they "
+    "have installed, like an IDE, a design app, or a DAW: these reveal their craft), then folders, plan tier, email "
+    "domain. Tune every task and the personal app to that profile; do not output the profile. "
     "Exactly 4 starters. Each title is 2-5 words. Each prompt is a concrete, safe, immediately runnable task "
     "referencing the user's real folders, file counts, or picked apps when possible; never invent facts, never "
     "propose deleting anything without review. The FIRST starter must be an audit sized for parallel sub-work "
@@ -49,8 +51,10 @@ P_SYSTEM = (
     "profile and habits (for example a daily morning brief, a weekly folder cleanup, a weekday summary of their "
     "connected apps). Each automation title is 2-4 words, prompt is one runnable instruction, cadence is exactly "
     "'daily', 'weekday', or 'weekly'. Automations must be safe to run unattended (never delete without review). "
-    "The greeting is one warm sentence that names 2-3 specific things "
-    "found on the machine. Never use em-dashes or en-dashes anywhere. No markdown, no commentary, JSON only."
+    "The greeting is one or two warm sentences: first say out loud, specifically and confidently, what this person is "
+    "into or working on (grounded in usage_summary, signal_apps, and folders, for example 'Looks like you live in "
+    "Xcode and ship iOS apps'), then name 2-3 concrete things you actually saw. Be specific, never generic, and never "
+    "name boring system apps. Never use em-dashes or en-dashes anywhere. No markdown, no commentary, JSON only."
 )
 
 

backend/tests/test_onboarding.py

@@ -7,7 +7,7 @@ from pathlib import Path
 import pytest
 
 from backend.apps.onboarding.identity import build_identity, decode_jwt_payload
-from backend.apps.onboarding.local_scan import run_local_scan
+from backend.apps.onboarding.local_scan import detect_signal_apps, run_local_scan
 from backend.apps.onboarding.models import PrepRequest, ScanResult
 from backend.apps.onboarding.prep import FALLBACK_STARTERS, build_prep, parse_prep
 from backend.apps.settings.models import AppSettings
@@ -78,6 +78,17 @@ def test_run_local_scan_counts_names_only(tmp_path: Path):
     assert "[user]" not in serialized
 
 
+def test_signal_apps_picks_tools_not_noise_and_avoids_substring_traps():
+    apps = ["Calculator", "Xcode", "Visual Studio Code", "Figma", "Search", "System Settings", "Adobe Photoshop 2024", "Arc"]
+    signal = detect_signal_apps(apps)
+    assert "Xcode" in signal and "Visual Studio Code" in signal and "Figma" in signal
+    assert "Adobe Photoshop 2024" in signal  # prefix match on a versioned name
+    assert "Arc" in signal
+    # "Search" contains "arc" as a substring but must NOT be treated as the Arc browser.
+    assert "Search" not in signal
+    assert "Calculator" not in signal and "System Settings" not in signal
+
+
 def test_parse_prep_strict_and_lenient():
     good = '{"greeting": "Hey!", "starters": [{"title": "A", "prompt": "do a"}, {"title": "B", "prompt": "do b"}]}'
     parsed = parse_prep(f"Sure! Here you go: {good}")