generated from coulomb/repo-seed
tegwick
70433cda61
detect/quality.py: is_real_coding_session drops health-checks / smoke-tests / interrupted / trivially-short sessions (event floor, repo present, substantive tool activity, non-trivial prompt). Wired into run_detect so signals only form over real sessions — fixes the abandoned false-positive. [detect.quality] knobs; existing detect/curate fixtures made realistic. 8 new tests; suite 80/80. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
76 lines
2.8 KiB
Python
76 lines
2.8 KiB
Python
"""Session-quality filter (T01).
|
|
|
|
The capture layer ingests *every* session it finds — including API health-checks,
|
|
smoke-tests, and interrupted runs (e.g. ``llm-connect`` firing "Say hello in one
|
|
word", or a transcript that is just ``[Request interrupted by user]``). These are
|
|
not real coding work, but the outcome heuristic labels the short ones ``abandoned``
|
|
and the clusterer then mints false-positive "problem" patterns from them.
|
|
|
|
:func:`is_real_coding_session` gates those out so Detect signals/clusters form only
|
|
over genuine coding sessions. It is intentionally conservative — a session counts
|
|
as real if it shows substantive activity, and is dropped only on clear trivial
|
|
markers. Thresholds come from ``[detect.quality]`` in ``config.toml``.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
from dataclasses import dataclass
|
|
from typing import Optional
|
|
|
|
# Prompt prefixes/markers that indicate a non-coding or interrupted session.
|
|
_TRIVIAL_PROMPTS = (
|
|
"say hello", "hello", "[request interrupted", "return only this json",
|
|
"ping", "ok", "<system-reminder>",
|
|
)
|
|
|
|
# Tool buckets that count as "substantive" coding activity.
|
|
_SUBSTANTIVE_TOOLS = (
|
|
"Edit", "Write", "Read", "Bash", "search_replace", "write", "read_file",
|
|
"run_terminal_command", "grep", "Grep", "glob", "Glob", "NotebookEdit",
|
|
)
|
|
|
|
|
|
@dataclass
|
|
class QualityConfig:
|
|
min_events: int = 20 # below this, not a real coding session
|
|
min_substantive: int = 3 # >= this many substantive tool calls required
|
|
min_prompt_len: int = 25 # first prompt shorter than this is suspect
|
|
|
|
|
|
def quality_config(config: Optional[dict] = None) -> QualityConfig:
|
|
d = (config or {}).get("detect", {}).get("quality", {}) if config else {}
|
|
return QualityConfig(
|
|
min_events=d.get("min_events", 20),
|
|
min_substantive=d.get("min_substantive", 3),
|
|
min_prompt_len=d.get("min_prompt_len", 25),
|
|
)
|
|
|
|
|
|
def _substantive_calls(digest: dict) -> int:
|
|
hist = digest.get("tool_histogram") or {}
|
|
return sum(n for t, n in hist.items() if t in _SUBSTANTIVE_TOOLS)
|
|
|
|
|
|
def is_real_coding_session(digest: dict, config: Optional[QualityConfig] = None) -> bool:
|
|
cfg = config or QualityConfig()
|
|
|
|
if not digest.get("repo"):
|
|
return False
|
|
if digest.get("event_count", 0) < cfg.min_events:
|
|
return False
|
|
if _substantive_calls(digest) < cfg.min_substantive:
|
|
return False
|
|
|
|
prompt = (digest.get("first_prompt") or "").strip().lower()
|
|
if len(prompt) < cfg.min_prompt_len:
|
|
return False
|
|
if any(prompt.startswith(p) for p in _TRIVIAL_PROMPTS):
|
|
return False
|
|
|
|
return True
|
|
|
|
|
|
def filter_real(digests: list[dict], config: Optional[QualityConfig] = None) -> list[dict]:
|
|
cfg = config or QualityConfig()
|
|
return [d for d in digests if is_real_coding_session(d, cfg)]
|