Files
mempalace/mempalace/convo_scanner.py
T
Igor Lins e Silva 035fe6d658 fix(llm): tighter refinement — word boundaries, JSON extraction, authoritative sources
Addresses issues found while reviewing the initial phase-2 implementation
against real data:

**Bug: uncertain bucket starved from the LLM.**
`discover_entities` was dropping the regex-uncertain bucket whenever real
git/manifest signal existed — which is exactly when `--llm` is most useful
for cleaning up prose noise. The uncertain candidates never reached the
refinement step. Fixed: only drop when `llm_provider is None`.

**Context collection: word boundaries, not substring.**
`_collect_contexts` used substring matching on lower-cased lines, so the
name "Go" matched "good", "going", "forgot". Switched to a
`(?<!\w)…(?!\w)` regex so short names only match at token boundaries.

**Authoritative-source detection replaces confidence threshold.**
Previously the refinement step skipped entries with `confidence >= 0.95`
to avoid second-guessing manifest-backed projects. That threshold was
fragile — the regex detector produces 0.99 confidence for things like
`code file reference (5x)` on framework names (OpenAPI, etc.), so those
skipped the LLM despite being regex-only noise. New helpers
`_is_authoritative_person` / `_is_authoritative_project` look at the
actual signal strings (commits, package.json, etc.) to decide.

**Now also refines regex-derived people.**
After #1148's high-pronoun-signal fix, the regex detector can promote
non-people to the `people` bucket (e.g. a capitalized common noun that
happened to appear near pronouns). The LLM now gets a chance to clean
those up, while git-authored people are still skipped.

**Robust JSON extraction.**
Small local models routinely wrap JSON output in prose ("Sure, here's
the classification: {…}"). The previous code-fence stripper failed on
that. `_extract_json_candidates` now does balanced-bracket extraction
with string-aware quote handling, so it recovers JSON from:
- raw responses
- markdown fenced blocks
- JSON embedded inside surrounding text
- multiple candidate objects/arrays

**Prompt guidance for frameworks vs user projects.**
Added an explicit instruction: frameworks, runtimes, APIs, cloud
services, and third-party vendors (Angular, OpenAPI, Terraform, Bun,
Google, etc.) are TOPIC unless the context clearly says it's the user's
own codebase. Directly addresses a false-positive pattern observed
during dev runs.

**Defensive mtime.**
`convo_scanner._safe_mtime` catches OSError during `stat()` — permission
changes, filesystem races, broken symlinks — and sorts the affected file
to the end of the newest-first order rather than crashing the scan.

**Cosmetic:** merged two adjacent f-strings on the same line in
`backends/chroma.py` and `llm_client.py` (no behaviour change).

15 new tests cover the OSError fallback, word-boundary matching, JSON
extraction variants, authoritative-source helpers, refining high-
confidence regex projects, and end-to-end LLM refinement preserving the
uncertain bucket.
2026-04-24 01:30:40 -03:00

161 lines
5.1 KiB
Python

"""
convo_scanner.py — Parse Claude Code conversation directories into ProjectInfo.
Claude Code stores sessions under ``~/.claude/projects/<slug>/<id>.jsonl``,
where the ``<slug>`` is the original CWD with ``/`` replaced by ``-``. That
encoding is lossy: we can't tell whether ``foo-bar`` in a slug is the
literal project name ``foo-bar`` or two path segments ``foo/bar``.
Fortunately, every message record in the JSONL carries a ``cwd`` field with
the true path. This scanner reads one record per session to recover the
accurate project name, falling back to slug-decoding only if the JSONL
is malformed or empty.
Output is the same ``ProjectInfo`` shape used by ``project_scanner``, so the
``discover_entities`` orchestrator can mix-and-match sources.
Public:
is_claude_projects_root(path) -> bool
scan_claude_projects(path) -> list[ProjectInfo]
"""
from __future__ import annotations
import json
from pathlib import Path
from typing import Optional
from mempalace.project_scanner import ProjectInfo
MAX_HEADER_LINES = 20 # lines to read per session looking for `cwd`
def is_claude_projects_root(path: Path) -> bool:
"""Return True if path looks like `.claude/projects/`.
Heuristic: at least one child dir whose name starts with ``-`` and which
contains at least one ``.jsonl`` file.
"""
if not path.is_dir():
return False
try:
children = list(path.iterdir())
except OSError:
return False
for child in children:
if not (child.is_dir() and child.name.startswith("-")):
continue
try:
if any(p.suffix == ".jsonl" for p in child.iterdir() if p.is_file()):
return True
except OSError:
continue
return False
def _extract_cwd_from_session(session_file: Path) -> Optional[str]:
"""Return the ``cwd`` from the first message record that carries one.
Returns None if the file can't be read, has no JSON, or no record has cwd.
"""
try:
with open(session_file, encoding="utf-8", errors="replace") as f:
for i, line in enumerate(f):
if i >= MAX_HEADER_LINES:
break
line = line.strip()
if not line:
continue
try:
obj = json.loads(line)
except json.JSONDecodeError:
continue
cwd = obj.get("cwd")
if isinstance(cwd, str) and cwd:
return cwd
except OSError:
return None
return None
def _decode_slug_fallback(slug: str) -> str:
"""Best-effort project name from slug when cwd is unavailable.
The slug is lossy (`/` and `-` both become `-`). Last non-empty segment
is the closest guess at the project name, preserving kebab-case is
impossible without cwd.
"""
stripped = slug.lstrip("-")
parts = [p for p in stripped.split("-") if p]
return parts[-1] if parts else slug
def _safe_mtime(path: Path) -> float:
"""Return file mtime, defaulting old on permission or filesystem errors."""
try:
return path.stat().st_mtime
except OSError:
return 0.0
def _resolve_project_name(project_dir: Path) -> str:
"""Read one session's cwd to recover the original project name.
Falls back to slug-decoding if no session has a readable cwd.
"""
sessions = sorted(
(p for p in project_dir.iterdir() if p.is_file() and p.suffix == ".jsonl"),
key=_safe_mtime,
reverse=True, # newest first — most likely to be well-formed
)
for session in sessions:
cwd = _extract_cwd_from_session(session)
if cwd:
return Path(cwd).name or cwd
return _decode_slug_fallback(project_dir.name)
def scan_claude_projects(path: str | Path) -> list[ProjectInfo]:
"""Scan a ``.claude/projects/`` directory for Claude Code conversations.
One ProjectInfo per subdir. ``has_git`` is False (the directory isn't a
repo itself) but ``total_commits`` is repurposed here as session count so
the UX surfaces a density signal for ranking.
"""
root = Path(path).expanduser().resolve()
if not is_claude_projects_root(root):
return []
projects: dict[str, ProjectInfo] = {}
for sub in sorted(root.iterdir()):
if not (sub.is_dir() and sub.name.startswith("-")):
continue
try:
sessions = [p for p in sub.iterdir() if p.is_file() and p.suffix == ".jsonl"]
except OSError:
continue
if not sessions:
continue
name = _resolve_project_name(sub)
session_count = len(sessions)
proj = ProjectInfo(
name=name,
repo_root=sub,
manifest=None,
has_git=False,
total_commits=session_count,
user_commits=session_count,
is_mine=True, # Claude Code sessions are authored by the user
)
existing = projects.get(name)
if existing is None or session_count > existing.user_commits:
projects[name] = proj
return sorted(
projects.values(),
key=lambda p: (-p.user_commits, p.name),
)