← Files Portable ResumeARCHIVED FILE
skills/.portable-resume/runtime/portable_resume/adapters/claude.py
66.2 KB · Oct 2, 2026 · 00:33 UTC
"""Read Claude Code JSONL sessions as inert context.
This adapter intentionally understands one pinned structural family only. It
does not invoke Claude Code, follow symlinks, or infer a transcript from an
unknown record shape.
"""
from __future__ import annotations
import hashlib
import json
import math
import os
import stat
import tempfile
import time
import uuid
from dataclasses import dataclass, field, replace
from datetime import datetime, timezone
from pathlib import Path
from typing import Any, Iterable, Mapping
from ..bounds import DEFAULT_BOUNDS, ReadBudget
from ..diagnostics import DiagnosticError
from ..model import Query, Session, SessionSummary, Turn
from ..paths import (
canonical_root,
canonicalize_cwd,
is_within,
require_regular_no_symlinks,
same_cwd,
)
from ..sanitize import sanitize_text, sanitize_turn_record
from ..snapshot import (
FileFingerprint,
FileSnapshot,
StableWindows,
snapshot_regular_file,
stable_read_windows,
stable_scan_tail_lines,
)
from .base import CapabilityReport, ResolvedRef
FORMAT_ID = "claude-jsonl-v1"
# Grok-build style: prefer cwd slug project dir; allow large ~/.claude/projects trees.
_PROJECT_DIR_LIMIT = 1_024
_METADATA_HEAD_BYTES = 4 * 1024 * 1024
_METADATA_TAIL_BYTES = 64 * 1024
_UUID_RECORD_TYPES = frozenset({"user", "assistant", "system"})
# Structural families we understand or safely ignore without payload interpretation.
# Unknown types are skipped with W_UNKNOWN_RECORD_SKIPPED (Grok resume-session parity).
_KNOWN_RECORD_TYPES = frozenset(
{
"user",
"assistant",
"system",
"summary",
"custom-title",
"ai-title",
"meta",
"queue-operation",
"last-prompt",
"tag",
"agent-name",
"agent-color",
"agent-setting",
"mode",
"permission-mode",
"worktree-state",
"progress",
"file-history-snapshot",
"file-history-delta",
"attribution-snapshot",
"content-replacement",
"context-collapse-commit",
"context-collapse-snapshot",
"attachment",
"bridge-session",
"pr-link",
}
)
_COMPACTION_SUBTYPES = frozenset({"compact_boundary", "compaction", "compact"})
_REPLAY_ENVELOPE_FIELDS = frozenset(
{"parentUuid", "cwd", "gitBranch", "slug", "promptId", "toolUseResult"}
)
def _slugify_cwd(cwd: str) -> str:
"""Match Claude Code project dir naming: non-alnum → '-' (same as Grok resume-session)."""
return "".join(char if char.isalnum() else "-" for char in cwd)
class _DuplicateKey(ValueError):
pass
def _object(pairs: list[tuple[str, Any]]) -> dict[str, Any]:
value: dict[str, Any] = {}
for key, item in pairs:
if key in value:
raise _DuplicateKey(key)
value[key] = item
return value
def _reject_json_constant(_value: str) -> Any:
raise ValueError("non-finite JSON number")
def _finite_json_float(value: str) -> float:
parsed = float(value)
if not math.isfinite(parsed):
raise ValueError("non-finite JSON number")
return parsed
@dataclass(slots=True)
class _TranscriptMetadata:
session_ids: set[str] = field(default_factory=set)
cwds: list[str] = field(default_factory=list)
custom_title: str | None = None
ai_title: str | None = None
summary: str | None = None
last_prompt: str | None = None
first_user: str | None = None
branch: str | None = None
created_at: str | None = None
records_seen: int = 0
def observe(self, record: Mapping[str, Any]) -> None:
self.records_seen += 1
session_id = record.get("sessionId")
if isinstance(session_id, str):
self.session_ids.add(session_id)
raw_cwd = record.get("cwd")
if isinstance(raw_cwd, str):
try:
cwd = canonicalize_cwd(raw_cwd)
except DiagnosticError as error:
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID) from error
if cwd not in self.cwds:
self.cwds.append(cwd)
if isinstance(record.get("customTitle"), str):
self.custom_title = record["customTitle"]
if isinstance(record.get("aiTitle"), str):
self.ai_title = record["aiTitle"]
if isinstance(record.get("summary"), str):
self.summary = record["summary"]
if isinstance(record.get("lastPrompt"), str):
self.last_prompt = record["lastPrompt"]
if isinstance(record.get("gitBranch"), str):
self.branch = record["gitBranch"]
stamp = _rfc3339(record.get("timestamp"))
if stamp is not None and (self.created_at is None or stamp < self.created_at):
self.created_at = stamp
if self.first_user is None and record.get("type") == "user" and not record.get("isMeta"):
message = record.get("message")
if isinstance(message, Mapping) and isinstance(message.get("content"), str):
candidate = message["content"]
if not candidate.lstrip().startswith("<command-name>"):
self.first_user = candidate
@property
def title(self) -> str | None:
return next(
(
item
for item in (
self.custom_title,
self.ai_title,
self.summary,
self.last_prompt,
self.first_user,
)
if item and item.strip()
),
None,
)
def selected_cwd(self, requested: str | None) -> str | None:
primary = self.cwds[0] if self.cwds else None
if requested is not None:
canonical = canonicalize_cwd(requested)
if primary is not None and same_cwd(canonical, primary):
return canonical
return None
return primary
@dataclass(frozen=True, slots=True)
class _TranscriptNode:
identifier: str
index: int
offset: int
digest: str
record: dict[str, Any]
@dataclass(slots=True)
class _TranscriptIndex:
metadata: _TranscriptMetadata
nodes: dict[str, _TranscriptNode]
bridge: dict[str, str | None]
warnings: tuple[str, ...]
def _root_candidate(query: Query) -> str:
if query.source_root:
return query.source_root
configured = os.environ.get("CLAUDE_CONFIG_DIR")
return configured if configured else os.path.expanduser("~/.claude")
def _existing_root(query: Query) -> str | None:
candidate = _root_candidate(query)
try:
if not os.path.isdir(candidate):
return None
return canonical_root(candidate)
except DiagnosticError:
if query.source_root:
raise
return None
def _regular_directory(path: str, root: str) -> bool:
try:
current = os.lstat(path)
except OSError:
return False
if stat.S_ISLNK(current.st_mode) or not stat.S_ISDIR(current.st_mode):
return False
canonical = canonicalize_cwd(path)
try:
return os.path.commonpath((canonical, root)) == root
except ValueError:
return False
def _bounded_names(directory: str, *, limit: int) -> list[str]:
values: list[str] = []
try:
with os.scandir(directory) as entries:
for entry in entries:
if len(values) >= limit:
raise DiagnosticError.limit_exceeded()
values.append(entry.name)
except DiagnosticError:
raise
except OSError as error:
raise DiagnosticError.source_busy(provider=FORMAT_ID) from error
values.sort()
return values
def _project_dirs(
root: str,
*,
prefer_slugs: tuple[str, ...] = (),
prefer_only: bool = False,
) -> list[str]:
"""Return project directories under ``<root>/projects``.
When *prefer_only* is true, only the preferred cwd-slug directories are
considered — the projects tree is not enumerated. This keeps exact UUID +
cwd discovery off the broad ``_PROJECT_DIR_LIMIT`` / scandir path.
"""
projects = os.path.join(root, "projects")
if not _regular_directory(projects, root):
return []
preferred: list[str] = []
seen: set[str] = set()
for slug in prefer_slugs:
if not slug or slug in seen:
continue
candidate = os.path.join(projects, slug)
if _regular_directory(candidate, root):
preferred.append(candidate)
seen.add(slug)
if prefer_only:
return preferred
names = _bounded_names(projects, limit=DEFAULT_BOUNDS.scanned_records)
others: list[str] = []
for name in names:
if name in seen:
continue
candidate = os.path.join(projects, name)
if _regular_directory(candidate, root):
others.append(candidate)
if len(preferred) + len(others) > _PROJECT_DIR_LIMIT:
raise DiagnosticError.limit_exceeded()
# Preferred (cwd slug) first — same discovery order as Grok resume-session.
return preferred + others
def _session_layout_ok(path: str, root: str) -> bool:
"""True when *path* is ``projects/<slug>/<uuid>.jsonl`` under *root*."""
if not is_within(path, root):
return False
relative = os.path.relpath(path, root)
parts = relative.split(os.sep)
if len(parts) != 3 or parts[0] != "projects" or not parts[1]:
return False
basename = parts[2]
if not basename.endswith(".jsonl"):
return False
try:
uuid.UUID(basename[:-6])
except ValueError:
return False
return True
def _regular_session_file(path: str) -> bool:
try:
current = os.lstat(path)
except OSError:
return False
return stat.S_ISREG(current.st_mode) and not stat.S_ISLNK(current.st_mode)
def _direct_uuid_under_slugs(
root: str,
exact_uuid: str,
prefer_slugs: tuple[str, ...],
) -> list[str]:
"""Return existing ``projects/<slug>/<uuid>.jsonl`` paths without tree scans."""
projects = os.path.join(root, "projects")
if not _regular_directory(projects, root):
return []
values: list[str] = []
seen: set[str] = set()
for slug in prefer_slugs:
if not slug or slug in seen:
continue
seen.add(slug)
project = os.path.join(projects, slug)
if not _regular_directory(project, root):
continue
candidate = os.path.join(project, f"{exact_uuid}.jsonl")
if _regular_session_file(candidate):
values.append(candidate)
return values
def _lexical_under_root(path: str, root: str) -> bool:
"""True when *path* is lexically under *root* (abspath spellings only)."""
try:
relative = os.path.relpath(os.path.abspath(path), os.path.abspath(root))
except ValueError:
return False
return relative != os.pardir and not relative.startswith(os.pardir + os.sep) and not os.path.isabs(relative)
def _lexical_session_shape(path: str, root: str) -> bool:
"""True when *path* is lexically ``projects/<slug>/<uuid>.jsonl`` under *root*."""
absolute = os.path.abspath(path)
abs_root = os.path.abspath(root)
if not _lexical_under_root(absolute, abs_root):
return False
try:
relative = os.path.relpath(absolute, abs_root)
except ValueError:
return False
parts = [part for part in Path(relative).parts if part not in ("", ".")]
if len(parts) != 3 or parts[0] != "projects" or not parts[1]:
return False
basename = parts[2]
if not basename.endswith(".jsonl"):
return False
try:
uuid.UUID(basename[:-6])
except ValueError:
return False
return True
def _missing_under_safe_parents(path: str, root: str) -> bool:
"""True when a missing *path* is a valid under-root session shape with safe parents.
Used to map ``E_UNSAFE_PATH`` from a missing leaf to ``E_NO_MATCH`` only when:
- the path is lexically ``projects/<slug>/<uuid>.jsonl`` under *root*
- the leaf does not exist
- every existing parent under the root is a real (non-symlink) directory
Symlinked parents, invalid layout, and non-ENOENT parent errors stay unsafe.
"""
absolute = os.path.abspath(path)
abs_root = os.path.abspath(root)
if not _lexical_session_shape(absolute, abs_root):
return False
if os.path.lexists(absolute):
return False
try:
relative = os.path.relpath(absolute, abs_root)
except ValueError:
return False
current = abs_root
parts = [part for part in Path(relative).parts if part not in ("", ".")]
for part in parts[:-1]:
if part == os.pardir:
return False
current = os.path.join(current, part)
try:
mode = os.lstat(current).st_mode
except FileNotFoundError:
# Intermediate component missing — under-root session shape no-match.
return True
except OSError:
# EACCES / other errors must not be reclassified as ordinary no-match.
return False
if stat.S_ISLNK(mode) or not stat.S_ISDIR(mode):
return False
return True
def _exact_path_candidate(root: str, query: Query) -> str | None:
"""Resolve an absolute path ref without store-wide discovery when possible.
Returns ``None`` when *query.ref* is not an absolute path. Raises
``E_NO_MATCH`` for a missing path that is still under the approved root
with safe real parents, and ``E_UNSAFE_PATH`` for outside/symlink/layout
failures (including a missing leaf under a symlinked parent).
"""
ref = query.ref.strip() if query.ref else None
if not ref or not os.path.isabs(ref):
return None
try:
path, _ = require_regular_no_symlinks(ref, root)
except DiagnosticError as error:
if error.code == "E_UNSAFE_PATH":
absolute = os.path.abspath(ref)
# Missing under-root target with safe parents → no match.
# Outside / symlink parents stay unsafe even when the leaf is absent.
if _missing_under_safe_parents(absolute, root):
raise DiagnosticError("E_NO_MATCH", source="claude", provider=FORMAT_ID) from error
raise
if not _session_layout_ok(path, root):
raise DiagnosticError.unsafe_path()
return path
def _paths_under_projects(
project_dirs: list[str],
*,
exact_uuid: str | None = None,
) -> list[str]:
"""Collect session JSONL paths under the given project directories.
Exact UUID lookups probe only ``<uuid>.jsonl`` per project (no session-dir
scandir). Non-exact list still bounds directory membership.
"""
values: list[str] = []
if exact_uuid is not None:
for project in project_dirs:
candidate = os.path.join(project, f"{exact_uuid}.jsonl")
if _regular_session_file(candidate):
values.append(candidate)
if len(values) > DEFAULT_BOUNDS.scanned_records:
raise DiagnosticError.limit_exceeded()
return values
for project in project_dirs:
names = _bounded_names(project, limit=DEFAULT_BOUNDS.scanned_records)
for name in names:
if not name.endswith(".jsonl"):
continue
stem = name[:-6]
try:
uuid.UUID(stem)
except ValueError:
continue
candidate = os.path.join(project, name)
if _regular_session_file(candidate):
values.append(candidate)
if len(values) > DEFAULT_BOUNDS.scanned_records:
raise DiagnosticError.limit_exceeded()
return values
def _session_paths(
root: str,
*,
prefer_slugs: tuple[str, ...] = (),
exact_uuid: str | None = None,
cwd_scoped: bool = False,
) -> list[str]:
"""Enumerate session JSONL paths.
Fast paths (issue #19):
- exact UUID + preferred cwd slug(s): construct
``projects/<slug>/<uuid>.jsonl`` and return it when present without
enumerating unrelated project directories
- cwd-scoped discovery uses preferred dirs only (no full projects scandir)
when those dirs exist
Exact UUID still falls back to a broader, basename-probed scan when the
direct candidate is absent. Recorded-cwd validation remains in
``_summary`` / ``show`` — slug presence alone is never authority.
"""
if exact_uuid is not None and prefer_slugs:
# Prefer direct slug path(s) first, then remaining project buckets via
# basename probe. Callers that can validate recorded cwd (list/show)
# should use `_exact_uuid_paths` so a cwd-mismatched direct file does
# not block an eligible relocated copy (Codex P2 on #19).
return _exact_uuid_paths(root, exact_uuid, prefer_slugs=prefer_slugs)
if cwd_scoped and prefer_slugs:
paths, _fallback = _cwd_scoped_session_paths(root, prefer_slugs)
return paths
project_dirs = _project_dirs(root, prefer_slugs=prefer_slugs, prefer_only=False)
return _paths_under_projects(project_dirs, exact_uuid=exact_uuid)
def _cwd_scoped_session_paths(
root: str,
prefer_slugs: tuple[str, ...],
) -> tuple[list[str], bool]:
preferred_dirs = _project_dirs(root, prefer_slugs=prefer_slugs, prefer_only=True)
if preferred_dirs:
return _paths_under_projects(preferred_dirs, exact_uuid=None), False
# Preferred slug dir missing: preserve legacy broad list so sessions whose
# project bucket name differs can still match recorded cwd.
project_dirs = _project_dirs(root, prefer_slugs=prefer_slugs, prefer_only=False)
return _paths_under_projects(project_dirs, exact_uuid=None), True
def _exact_uuid_paths(
root: str,
exact_uuid: str,
*,
prefer_slugs: tuple[str, ...] = (),
) -> list[str]:
"""Ordered exact-UUID candidates: preferred slug files, then broad probe.
Preferred direct files come first so callers can accept them without a full
projects scandir when recorded-cwd validation succeeds. Broad basename
probing still runs so a cwd-mismatched file under the guessed slug cannot
hide an eligible copy in another project bucket.
"""
direct = _direct_uuid_under_slugs(root, exact_uuid, prefer_slugs)
project_dirs = _project_dirs(root, prefer_slugs=prefer_slugs, prefer_only=False)
broad = _paths_under_projects(project_dirs, exact_uuid=exact_uuid)
seen: set[str] = set()
ordered: list[str] = []
for path in (*direct, *broad):
if path in seen:
continue
seen.add(path)
ordered.append(path)
return ordered
def _prefer_slugs_for(query: Query) -> tuple[str, ...]:
if not query.cwd:
return ()
try:
return (_slugify_cwd(canonicalize_cwd(query.cwd)),)
except DiagnosticError:
return (_slugify_cwd(query.cwd),)
def _exact_uuid_ref(value: str | None) -> str | None:
if not value:
return None
try:
return str(uuid.UUID(value))
except ValueError:
return None
def _decode_record(raw: bytes, *, terminal_partial: bool) -> tuple[dict[str, Any] | None, str | None]:
stripped = raw.strip()
if not stripped:
return None, None
try:
text = stripped.decode("utf-8")
value = json.loads(
text,
object_pairs_hook=_object,
parse_constant=_reject_json_constant,
parse_float=_finite_json_float,
)
except (UnicodeDecodeError, ValueError, _DuplicateKey, RecursionError) as error:
if terminal_partial:
return None, "W_PARTIAL_TAIL"
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID) from error
if not isinstance(value, dict) or not isinstance(value.get("type"), str):
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID)
if value["type"] not in _KNOWN_RECORD_TYPES:
return None, "W_UNKNOWN_RECORD_SKIPPED"
return value, None
def _parse_lines(data: bytes, budget: ReadBudget) -> tuple[list[dict[str, Any]], tuple[str, ...]]:
"""Parse a small in-memory Claude fixture; full transcripts use streaming."""
records: list[dict[str, Any]] = []
warnings: list[str] = []
lines = data.splitlines(keepends=True)
for index, raw in enumerate(lines):
budget.consume_records()
terminal_partial = index == len(lines) - 1 and not raw.endswith((b"\n", b"\r"))
value, warning = _decode_record(raw, terminal_partial=terminal_partial)
if warning is not None:
warnings.append(warning)
if value is None:
if terminal_partial and warning == "W_PARTIAL_TAIL":
break
continue
records.append(value)
if not records:
raise DiagnosticError("E_UNSUPPORTED_FORMAT", source="claude", provider=FORMAT_ID)
return records, tuple(dict.fromkeys(warnings))
def _scan_metadata_chunk(
data: bytes,
*,
budget: ReadBudget,
metadata: _TranscriptMetadata,
warnings: list[str],
starts_mid_line: bool,
ends_at_eof: bool,
stop_when_primary_ready: bool,
stop_when_cwd_ready: bool = False,
) -> None:
lines = data.splitlines(keepends=True)
start = 1 if starts_mid_line and lines else 0
for index in range(start, len(lines)):
raw = lines[index]
is_last = index == len(lines) - 1
has_terminator = raw.endswith((b"\n", b"\r"))
if is_last and not has_terminator and not ends_at_eof:
break
budget.consume_records()
value, warning = _decode_record(
raw,
terminal_partial=is_last and not has_terminator and ends_at_eof,
)
if warning is not None:
warnings.append(warning)
if value is None:
if warning == "W_PARTIAL_TAIL":
break
continue
metadata.observe(value)
if stop_when_cwd_ready and metadata.cwds:
break
if (
stop_when_primary_ready
and metadata.cwds
and metadata.created_at is not None
and metadata.title is not None
):
break
def _metadata_windows(
path: str,
root: str,
budget: ReadBudget,
*,
cwd_only: bool = False,
) -> tuple[StableWindows, _TranscriptMetadata, tuple[str, ...]]:
_validate_claude_bounds(budget)
remaining = max(
0,
min(budget.limits.source_read_bytes, DEFAULT_BOUNDS.source_read_bytes)
- budget.bytes_read,
)
head_bytes = min(_METADATA_HEAD_BYTES, remaining)
tail_bytes = min(_METADATA_TAIL_BYTES, max(0, remaining - head_bytes))
observation = stable_read_windows(
path,
root=root,
head_bytes=head_bytes,
tail_bytes=tail_bytes,
max_bytes=min(budget.limits.source_read_bytes, DEFAULT_BOUNDS.source_read_bytes),
attempts=min(budget.limits.snapshot_attempts, DEFAULT_BOUNDS.snapshot_attempts),
membership_limit=min(budget.limits.scanned_records, DEFAULT_BOUNDS.scanned_records),
budget=budget,
require_size_within_max=False,
)
metadata = _TranscriptMetadata()
warnings: list[str] = []
# Compare against the remaining byte budget, not the internal sample size
# (a healthy 17 MiB session under 256 MiB must not be flagged).
if observation.fingerprint.size > remaining:
warnings.append("W_TRUNCATED")
full_in_head = observation.fingerprint.size <= len(observation.head)
_scan_metadata_chunk(
observation.head,
budget=budget,
metadata=metadata,
warnings=warnings,
starts_mid_line=False,
ends_at_eof=full_in_head,
stop_when_primary_ready=not full_in_head and not cwd_only,
stop_when_cwd_ready=cwd_only,
)
if not full_in_head and observation.tail and not (cwd_only and metadata.cwds):
tail = observation.tail
starts_mid_line = observation.tail_offset > 0
if observation.tail_offset < len(observation.head):
overlap = len(observation.head) - observation.tail_offset
tail = tail[min(overlap, len(tail)) :]
starts_mid_line = bool(tail)
_scan_metadata_chunk(
tail,
budget=budget,
metadata=metadata,
warnings=warnings,
starts_mid_line=starts_mid_line,
ends_at_eof=True,
stop_when_primary_ready=False,
stop_when_cwd_ready=cwd_only,
)
if metadata.records_seen == 0:
# Precharged/zero byte budget: keep the limit diagnostic. Missing
# sampled metadata under a healthy budget stays E_UNSUPPORTED_FORMAT.
if remaining <= 0 or observation.fingerprint.size > remaining:
raise DiagnosticError("E_LIMIT_EXCEEDED", source="claude", provider=FORMAT_ID)
raise DiagnosticError("E_UNSUPPORTED_FORMAT", source="claude", provider=FORMAT_ID)
return observation, metadata, tuple(dict.fromkeys(warnings))
def _recorded_cwd_matches(
path: str,
root: str,
requested_cwd: str,
budget: ReadBudget,
) -> bool:
_observation, metadata, _warnings = _metadata_windows(
path,
root,
budget,
cwd_only=True,
)
return metadata.selected_cwd(requested_cwd) is not None
def _provisional_metadata_budget(budget: ReadBudget) -> ReadBudget:
return ReadBudget(
budget.limits,
records=budget.records,
transcript_records_read=budget.transcript_records_read,
bytes_read=budget.bytes_read,
turns=budget.turns,
)
def _commit_metadata_budget(target: ReadBudget, admitted: ReadBudget) -> None:
target.consume_records(admitted.records - target.records)
target.consume_bytes(admitted.bytes_read - target.bytes_read)
def _validate_claude_bounds(budget: ReadBudget) -> None:
limits = budget.limits
bounded = (
("scanned_records", 0, DEFAULT_BOUNDS.scanned_records),
("transcript_records", 0, DEFAULT_BOUNDS.transcript_records),
("record_bytes", 0, DEFAULT_BOUNDS.record_bytes),
("source_read_bytes", 0, DEFAULT_BOUNDS.source_read_bytes),
("snapshot_attempts", 1, DEFAULT_BOUNDS.snapshot_attempts),
)
if any(not minimum <= getattr(limits, name) <= maximum for name, minimum, maximum in bounded):
raise DiagnosticError.invalid(source="claude")
def _rfc3339(value: object) -> str | None:
if not isinstance(value, str):
return None
try:
parsed = datetime.fromisoformat(value.replace("Z", "+00:00"))
except ValueError:
return None
if parsed.tzinfo is None:
parsed = parsed.replace(tzinfo=timezone.utc)
return parsed.astimezone(timezone.utc).isoformat(timespec="microseconds").replace("+00:00", "Z")
def _mtime(read: StableWindows | FileSnapshot) -> str:
return datetime.fromtimestamp(read.fingerprint.mtime_ns / 1_000_000_000, timezone.utc).isoformat(
timespec="microseconds"
).replace("+00:00", "Z")
def _within(updated_at: str | None, minutes: int | None) -> bool:
from .common import within_age
return within_age(updated_at, minutes, default_minutes=DEFAULT_BOUNDS.listing_age_minutes)
def _record_cwd(records: Iterable[Mapping[str, Any]], requested: str | None = None) -> str | None:
metadata = _TranscriptMetadata()
for record in records:
metadata.observe(record)
return metadata.selected_cwd(requested)
def _session_id(path: str, records: Iterable[Mapping[str, Any]]) -> str:
file_id = Path(path).stem
values = {item for record in records if isinstance((item := record.get("sessionId")), str)}
if len(values) > 1 or (values and next(iter(values)) != file_id):
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID)
return file_id
def _title_and_branch(records: Iterable[Mapping[str, Any]]) -> tuple[str | None, str | None]:
custom: str | None = None
ai: str | None = None
summary: str | None = None
last_prompt: str | None = None
first_user: str | None = None
branch: str | None = None
for record in records:
if isinstance(record.get("customTitle"), str):
custom = record["customTitle"]
if isinstance(record.get("aiTitle"), str):
ai = record["aiTitle"]
if isinstance(record.get("summary"), str):
summary = record["summary"]
if isinstance(record.get("lastPrompt"), str):
last_prompt = record["lastPrompt"]
if isinstance(record.get("gitBranch"), str):
branch = record["gitBranch"]
if first_user is None and record.get("type") == "user" and not record.get("isMeta"):
message = record.get("message")
if isinstance(message, Mapping) and isinstance(message.get("content"), str):
candidate = message["content"]
if not candidate.lstrip().startswith("<command-name>"):
first_user = candidate
return next((item for item in (custom, ai, summary, last_prompt, first_user) if item and item.strip()), None), branch
def _parent_bridge_map(
records: list[dict[str, Any]],
) -> tuple[dict[str, str | None], dict[str, str]]:
"""Map every uuid → parentUuid for hop-through non-conversation nodes.
Live Claude transcripts often parent user/assistant rows to *attachment*
(hooks, skill listings, …). Those UUIDs are not conversation-renderable but
must not break the walk (Grok-style main-line recovery).
"""
bridge: dict[str, str | None] = {}
digests: dict[str, str] = {}
for record in records:
_observe_replay_record(record, digests=digests, bridge=bridge)
return bridge, digests
def _replay_digest(record: Mapping[str, Any]) -> str:
semantic = {key: value for key, value in record.items() if key not in _REPLAY_ENVELOPE_FIELDS}
message = semantic.get("message")
if isinstance(message, Mapping):
# Claude replay/bridge rows may zero usage accounting while preserving
# the actual role/content/model/stop semantics of the message.
semantic["message"] = {key: value for key, value in message.items() if key != "usage"}
try:
encoded = json.dumps(
semantic,
ensure_ascii=False,
sort_keys=True,
separators=(",", ":"),
).encode("utf-8")
except (TypeError, ValueError, RecursionError) as error:
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID) from error
return hashlib.sha256(encoded).hexdigest()
def _observe_replay_record(
record: Mapping[str, Any],
*,
digests: dict[str, str],
bridge: dict[str, str | None],
) -> tuple[str, str] | None:
identifier = record.get("uuid")
if identifier is None:
if record.get("type") in _UUID_RECORD_TYPES:
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID)
return None
if not isinstance(identifier, str):
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID)
try:
uuid.UUID(identifier)
except ValueError as error:
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID) from error
digest = _replay_digest(record)
if identifier in digests and digests[identifier] != digest:
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID)
digests[identifier] = digest
parent = record.get("parentUuid")
if parent is not None:
if not isinstance(parent, str):
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID)
try:
uuid.UUID(parent)
except ValueError as error:
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID) from error
bridge[identifier] = parent
logical = record.get("logicalParentUuid")
if logical is not None:
if not isinstance(logical, str):
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID)
try:
uuid.UUID(logical)
except ValueError as error:
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID) from error
return identifier, digest
def _resolve_parent_id(
parent: object,
*,
nodes: Mapping[str, tuple[int, dict[str, Any]]],
bridge: Mapping[str, str | None],
record: Mapping[str, Any],
) -> tuple[str | None, bool]:
"""Return (next_conversation_or_missing_id, broken).
When *parent* is a non-conversation uuid present only in *bridge*, hop
until a conversation node, a null parent, a cycle, or a truly missing id.
"""
if parent is None:
return None, False
if not isinstance(parent, str):
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID)
hop_seen: set[str] = set()
current: str | None = parent
while current is not None:
if current in nodes:
return current, False
if current in hop_seen:
return current, True
hop_seen.add(current)
if current not in bridge:
# Optional logicalParentUuid on the *current conversation record*
# only applies once at the start; after hops we only use bridge.
if current == parent:
logical = record.get("logicalParentUuid")
if isinstance(logical, str) and logical and logical != parent:
current = logical
continue
return current, True
current = bridge[current]
return None, False
def _lineage_ids(
nodes: Mapping[str, tuple[int, dict[str, Any]]],
bridge: Mapping[str, str | None],
) -> tuple[list[str], tuple[str, ...]]:
if not nodes:
raise DiagnosticError("E_UNSUPPORTED_FORMAT", source="claude", provider=FORMAT_ID)
# Leaf = conversation node never referenced as a *resolved* parent of another
# conversation node (bridge hops through attachment/etc. first).
parent_ids: set[str] = set()
for _identifier, (_index, record) in nodes.items():
parent = record.get("parentUuid")
resolved, _ = _resolve_parent_id(parent, nodes=nodes, bridge=bridge, record=record)
if resolved is not None:
parent_ids.add(resolved)
elif parent is None:
logical = record.get("logicalParentUuid")
resolved_logical, _ = _resolve_parent_id(
logical, nodes=nodes, bridge=bridge, record=record
)
if resolved_logical is not None:
parent_ids.add(resolved_logical)
leaves = [(index, identifier, record) for identifier, (index, record) in nodes.items() if identifier not in parent_ids]
if not leaves:
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID)
_, current_id, _ = max(
leaves,
key=lambda item: (_rfc3339(item[2].get("timestamp")) or "", item[0], item[1]),
)
reverse: list[str] = []
seen: set[str] = set()
warnings: list[str] = []
while current_id:
if current_id in seen:
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID)
seen.add(current_id)
found = nodes.get(current_id)
if found is None:
warnings.append("W_BROKEN_CHAIN")
break
record = found[1]
if record.get("type") == "system" and record.get("subtype") in _COMPACTION_SUBTYPES:
break
reverse.append(current_id)
parent = record.get("parentUuid")
if parent is None and isinstance(record.get("logicalParentUuid"), str):
parent = record.get("logicalParentUuid")
if parent is None:
break
next_id, broken = _resolve_parent_id(parent, nodes=nodes, bridge=bridge, record=record)
if broken:
warnings.append("W_BROKEN_CHAIN")
break
if next_id is None:
break
current_id = next_id
reverse.reverse()
return reverse, tuple(dict.fromkeys(warnings))
def _logical_lineage(records: list[dict[str, Any]]) -> tuple[list[dict[str, Any]], tuple[str, ...]]:
nodes: dict[str, tuple[int, dict[str, Any]]] = {}
bridge, _digests = _parent_bridge_map(records)
for index, record in enumerate(records):
if record.get("type") not in _UUID_RECORD_TYPES:
continue
identifier = record.get("uuid")
if not isinstance(identifier, str):
continue
try:
uuid.UUID(identifier)
except ValueError as error:
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID) from error
if record.get("isSidechain") is True:
continue
# Claude bridge/replay blocks may repeat a semantic conversation node
# with a new parent and envelope metadata. The latest physical version
# is the authoritative graph edge; core-content conflicts still fail.
nodes[identifier] = (index, record)
identifiers, warnings = _lineage_ids(nodes, bridge)
return [nodes[identifier][1] for identifier in identifiers], warnings
def _flatten_text(value: object) -> str | None:
if isinstance(value, str):
return value
if isinstance(value, list):
chunks: list[str] = []
for item in value:
if isinstance(item, str):
chunks.append(item)
elif isinstance(item, Mapping) and item.get("type") in {"text", "input_text", "output_text"}:
if isinstance(item.get("text"), str):
chunks.append(item["text"])
return "\n".join(chunks) if chunks else None
return None
@dataclass(frozen=True, slots=True)
class _PendingToolCall:
name: str | None
rendered_input: str
timestamp: str | None
input_truncated: bool
@dataclass(slots=True)
class _ToolCallContext:
maximum_pending: int
maximum_chars: int
pending: dict[str, _PendingToolCall] = field(default_factory=dict)
warnings: list[str] = field(default_factory=list)
def observe_call(self, item: Mapping[str, Any], timestamp: str | None) -> None:
identifier = item.get("id")
if not isinstance(identifier, str) or not identifier:
return
if identifier not in self.pending and len(self.pending) >= self.maximum_pending:
return
serialized = json.dumps(
item.get("input"),
ensure_ascii=False,
separators=(",", ":"),
sort_keys=True,
)
cleaned = sanitize_text(serialized, max_chars=self._input_allowance())
self.warnings.extend(cleaned.warnings)
name = item.get("name")
self.pending[identifier] = _PendingToolCall(
name=name if isinstance(name, str) else None,
rendered_input=cleaned.text,
timestamp=timestamp,
input_truncated=cleaned.truncated,
)
def correlated_result(
self,
item: Mapping[str, Any],
result: str,
timestamp: str | None,
) -> dict[str, Any]:
identifier = item.get("tool_use_id")
call = self.pending.pop(identifier, None) if isinstance(identifier, str) else None
if call is None:
return {"role": "tool", "content": result, "timestamp": timestamp}
content, result_truncated = self._combined_content(call.rendered_input, result)
return {
"role": "tool",
"content": content,
"tool_name": call.name,
"timestamp": timestamp,
"_pretruncated": call.input_truncated or result_truncated,
}
def missing_results(self) -> list[dict[str, Any]]:
values: list[dict[str, Any]] = []
for call in self.pending.values():
content, result_truncated = self._combined_content(
call.rendered_input,
"[missing tool result]",
)
values.append(
{
"role": "tool",
"content": content,
"tool_name": call.name,
"timestamp": call.timestamp,
"_pretruncated": call.input_truncated or result_truncated,
}
)
self.pending.clear()
return values
def _input_allowance(self) -> int:
framing = len("Input: \nResult:\n")
result_reserve = self.maximum_chars // 2
return max(
0,
min(
self.maximum_chars // 4,
self.maximum_chars - framing - result_reserve,
),
)
def _combined_content(self, rendered_input: str, result: str) -> tuple[str, bool]:
prefix = f"Input: {rendered_input}\nResult:\n"
if len(prefix) >= self.maximum_chars:
cleaned = sanitize_text(result, max_chars=self.maximum_chars)
self.warnings.extend(cleaned.warnings)
return cleaned.text, cleaned.truncated
cleaned = sanitize_text(result, max_chars=self.maximum_chars - len(prefix))
self.warnings.extend(cleaned.warnings)
return prefix + cleaned.text, cleaned.truncated
def _turn_records(
record: Mapping[str, Any],
tool_calls: _ToolCallContext,
) -> list[dict[str, Any]]:
record_type = record.get("type")
if record_type not in {"user", "assistant"} or record.get("isMeta") is True:
return []
message = record.get("message")
if not isinstance(message, Mapping):
return []
message_role = message.get("role") if isinstance(message.get("role"), str) else record_type
if message_role not in {"user", "assistant"}:
return []
timestamp = _rfc3339(record.get("timestamp"))
content = message.get("content")
if isinstance(content, str):
return [{"role": message_role, "content": content, "timestamp": timestamp}]
if not isinstance(content, list):
return []
values: list[dict[str, Any]] = []
for item in content:
if not isinstance(item, Mapping):
continue
kind = item.get("type")
if kind in {"thinking", "redacted_thinking", "signature"}:
continue
if kind == "tool_use":
tool_calls.observe_call(item, timestamp)
continue
if kind in {"text", "input_text", "output_text"} and isinstance(item.get("text"), str):
values.append({"role": message_role, "content": item["text"], "timestamp": timestamp})
elif kind == "tool_result":
text = _flatten_text(item.get("content"))
if text is not None:
values.append(tool_calls.correlated_result(item, text, timestamp))
return values
def _metadata_session_id(path: str, metadata: _TranscriptMetadata) -> str:
file_id = Path(path).stem
if len(metadata.session_ids) > 1 or (
metadata.session_ids and next(iter(metadata.session_ids)) != file_id
):
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID)
return file_id
def _graph_record(record: Mapping[str, Any]) -> dict[str, Any]:
return {
key: record[key]
for key in (
"type",
"uuid",
"parentUuid",
"logicalParentUuid",
"timestamp",
"subtype",
"isSidechain",
)
if key in record
}
def _index_snapshot(snapshot: FileSnapshot, budget: ReadBudget) -> _TranscriptIndex:
_validate_claude_bounds(budget)
metadata = _TranscriptMetadata()
nodes: dict[str, _TranscriptNode] = {}
bridge: dict[str, str | None] = {}
digests: dict[str, str] = {}
warnings: list[str] = []
maximum_record = budget.limits.record_bytes
with open(snapshot.path, "rb") as handle:
index = 0
while True:
offset = handle.tell()
raw = handle.readline(maximum_record + 1)
if not raw:
break
budget.consume_transcript_records()
if len(raw) > maximum_record:
raise DiagnosticError.limit_exceeded()
terminal_partial = not raw.endswith((b"\n", b"\r"))
record, warning = _decode_record(raw, terminal_partial=terminal_partial)
if warning is not None:
warnings.append(warning)
if record is None:
if warning == "W_PARTIAL_TAIL":
break
index += 1
continue
metadata.observe(record)
observed = _observe_replay_record(record, digests=digests, bridge=bridge)
if observed is not None and record.get("type") in _UUID_RECORD_TYPES:
identifier, digest = observed
if record.get("isSidechain") is not True:
nodes[identifier] = _TranscriptNode(
identifier=identifier,
index=index,
offset=offset,
digest=digest,
record=_graph_record(record),
)
index += 1
if metadata.records_seen == 0 or not nodes:
raise DiagnosticError("E_UNSUPPORTED_FORMAT", source="claude", provider=FORMAT_ID)
return _TranscriptIndex(
metadata=metadata,
nodes=nodes,
bridge=bridge,
warnings=tuple(dict.fromkeys(warnings)),
)
def _indexed_lineage(index: _TranscriptIndex) -> tuple[list[_TranscriptNode], tuple[str, ...]]:
graph = {
identifier: (node.index, node.record)
for identifier, node in index.nodes.items()
}
identifiers, warnings = _lineage_ids(graph, index.bridge)
return [index.nodes[identifier] for identifier in identifiers], warnings
def _load_lineage_records(
snapshot: FileSnapshot,
nodes: Iterable[_TranscriptNode],
*,
maximum_record: int,
) -> list[dict[str, Any]]:
records: list[dict[str, Any]] = []
with open(snapshot.path, "rb") as handle:
for node in nodes:
handle.seek(node.offset)
raw = handle.readline(maximum_record + 1)
if not raw or len(raw) > maximum_record:
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID)
record, warning = _decode_record(raw, terminal_partial=False)
if warning is not None or record is None:
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID)
if (
record.get("uuid") != node.identifier
or _replay_digest(record) != node.digest
or _graph_record(record) != node.record
):
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID)
records.append(record)
return records
def _summary(path: str, root: str, query: Query, budget: ReadBudget) -> SessionSummary | None:
observation, metadata, warnings = _metadata_windows(path, root, budget)
identifier = _metadata_session_id(path, metadata)
cwd = metadata.selected_cwd(query.cwd)
if query.cwd is not None and (cwd is None or not same_cwd(cwd, query.cwd)):
return None
updated = _mtime(observation)
if not _within(updated, query.within_min):
return None
# Title may be null; still list coherent sessions so exact-id/latest remain selectable.
return SessionSummary(
source="claude",
session_id=identifier,
source_path=path,
title=metadata.title,
cwd=cwd,
branch=metadata.branch,
created_at=metadata.created_at,
updated_at=updated,
provider=FORMAT_ID,
warnings=warnings,
)
def _assemble_from_index(
observation: FileSnapshot,
index: _TranscriptIndex,
path: str,
ref: ResolvedRef,
query: Query,
budget: ReadBudget,
) -> Session:
if _metadata_session_id(path, index.metadata) != ref.session_id:
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID)
cwd = index.metadata.selected_cwd(query.cwd)
if query.cwd is not None and cwd is None:
raise DiagnosticError("E_NO_MATCH", source="claude", provider=FORMAT_ID)
lineage, lineage_warnings = _indexed_lineage(index)
records = _load_lineage_records(
observation,
lineage,
maximum_record=budget.limits.record_bytes,
)
turns: list[Turn] = []
all_warnings = list((*index.warnings, *lineage_warnings))
turn_bounds = replace(DEFAULT_BOUNDS, tool_output_chars=query.max_tool_chars)
tool_calls = _ToolCallContext(
maximum_pending=min(
budget.limits.scanned_records,
DEFAULT_BOUNDS.scanned_records,
),
maximum_chars=query.max_tool_chars,
)
def append_turn(raw: Mapping[str, Any]) -> None:
turn, turn_warnings = sanitize_turn_record(
raw,
ordinal=len(turns),
bounds=turn_bounds,
)
all_warnings.extend(turn_warnings)
if turn is not None:
if raw.get("_pretruncated") is True and not turn.truncated:
turn = replace(turn, truncated=True)
budget.consume_turns()
turns.append(turn)
for record in records:
for raw in _turn_records(record, tool_calls):
append_turn(raw)
for raw in tool_calls.missing_results():
append_turn(raw)
all_warnings.extend(tool_calls.warnings)
last_user = next(
(turn.content for turn in reversed(turns) if turn.role == "user"),
None,
)
last_assistant = next(
(turn.content for turn in reversed(turns) if turn.role == "assistant"),
None,
)
return Session(
source="claude",
session_id=ref.session_id,
source_path=path,
title=index.metadata.title,
cwd=cwd,
branch=index.metadata.branch,
created_at=index.metadata.created_at,
updated_at=_mtime(observation),
last_user_request=last_user,
last_assistant_action=last_assistant,
turns=tuple(turns),
warnings=tuple(dict.fromkeys(all_warnings)),
)
def _build_tail_graph(
path: str,
root: str,
budget: ReadBudget,
) -> tuple[FileSnapshot, _TranscriptIndex]:
"""Soft-degraded suffix graph for oversized sessions (#258).
Called only after the full snapshot/index path raised E_LIMIT_EXCEEDED.
Admits the stable tail window under remaining ``source_read_bytes`` /
``transcript_records``, mirrors the admitted lines into a private temp
file (bounded), indexes them, and returns a synthetic FileSnapshot so the
shared lineage/turn assembly can reuse ``_load_lineage_records``.
"""
_validate_claude_bounds(budget)
maximum_record = min(budget.limits.record_bytes, DEFAULT_BOUNDS.record_bytes)
try:
before = os.lstat(path)
except OSError as error:
raise DiagnosticError.source_busy(provider=FORMAT_ID) from error
temporary = tempfile.TemporaryDirectory(prefix="portable-resume-claude-tail-")
try:
target = Path(temporary.name) / "tail.jsonl"
metadata = _TranscriptMetadata()
nodes: dict[str, _TranscriptNode] = {}
bridge: dict[str, str | None] = {}
digests: dict[str, str] = {}
warnings: list[str] = ["W_TRUNCATED"]
index = 0
remaining = max(
0,
min(budget.limits.source_read_bytes, DEFAULT_BOUNDS.source_read_bytes)
- budget.bytes_read,
)
if before.st_size <= remaining:
head_bytes = min(_METADATA_HEAD_BYTES, maximum_record, remaining)
else:
head_bytes = min(
_METADATA_HEAD_BYTES,
maximum_record,
max(0, remaining - min(maximum_record, remaining)),
)
def observe_stable_head(data: bytes, complete: bool) -> None:
_scan_metadata_chunk(
data,
budget=budget,
metadata=metadata,
warnings=warnings,
starts_mid_line=False,
ends_at_eof=complete,
stop_when_primary_ready=False,
stop_when_cwd_ready=True,
)
lines = stable_scan_tail_lines(
path,
root=root,
budget=budget,
charge_transcript=True,
max_line_bytes=maximum_record,
stable_head_bytes=head_bytes,
on_stable_head=observe_stable_head,
)
with open(target, "wb") as handle:
for line in lines:
if not line.utf8_valid:
# Terminal partial UTF-8: mirror fast-path W_PARTIAL_TAIL
# (re-encoding U+FFFD would exceed record_bytes).
if not line.terminated:
warnings.append("W_PARTIAL_TAIL")
break
raise DiagnosticError("E_CORRUPT_RECORD", source="claude", provider=FORMAT_ID)
offset = handle.tell()
raw = line.text.encode("utf-8")
if line.terminated:
raw += b"\n"
handle.write(raw)
# ORIGINAL physical length (incl. CRLF) cannot exceed record_bytes;
# the LF-reconstructed mirror would otherwise shrink CRLF records.
if len(raw) + (1 if line.crlf else 0) > maximum_record:
raise DiagnosticError.limit_exceeded()
record, warning = _decode_record(
raw, terminal_partial=not line.terminated
)
if warning is not None:
warnings.append(warning)
if record is None:
if warning == "W_PARTIAL_TAIL":
break
index += 1
continue
metadata.observe(record)
observed = _observe_replay_record(record, digests=digests, bridge=bridge)
if observed is not None and record.get("type") in _UUID_RECORD_TYPES:
identifier, digest = observed
if record.get("isSidechain") is not True:
nodes[identifier] = _TranscriptNode(
identifier=identifier,
index=index,
offset=offset,
digest=digest,
record=_graph_record(record),
)
index += 1
try:
after = os.lstat(path)
except OSError as error:
raise DiagnosticError.source_busy(provider=FORMAT_ID) from error
if (
before.st_dev != after.st_dev
or before.st_ino != after.st_ino
or before.st_mtime_ns != after.st_mtime_ns
or before.st_size != after.st_size
):
raise DiagnosticError.source_busy(provider=FORMAT_ID)
if metadata.records_seen == 0 or not nodes:
# Fallback runs only after the full path raised E_LIMIT_EXCEEDED.
raise DiagnosticError("E_LIMIT_EXCEEDED", source="claude", provider=FORMAT_ID)
fingerprint = FileFingerprint(
before.st_dev,
before.st_ino,
before.st_mode,
before.st_size,
before.st_mtime_ns,
)
observation = FileSnapshot(
directory=temporary.name,
path=str(target),
source_name=os.path.basename(path),
fingerprint=fingerprint,
attempts=1,
_temporary=temporary,
)
return observation, _TranscriptIndex(
metadata=metadata,
nodes=nodes,
bridge=bridge,
warnings=tuple(dict.fromkeys(warnings)),
)
except BaseException:
temporary.cleanup()
raise
class ClaudeAdapter:
key = "claude"
def approved_roots(self, query: Query) -> tuple[str, ...]:
root = _existing_root(query)
return (root,) if root else ()
def probe(self, query: Query) -> CapabilityReport:
try:
root = _existing_root(query)
if root is None:
return CapabilityReport(self.key, FORMAT_ID, "unavailable")
def _probe_path(path: str) -> CapabilityReport | None:
try:
_observation, metadata, _warnings = _metadata_windows(
path, root, ReadBudget()
)
_metadata_session_id(path, metadata)
return CapabilityReport(
self.key, FORMAT_ID, "supported", root=root, evidence=(FORMAT_ID,)
)
except DiagnosticError as error:
if error.code in {"E_UNSAFE_PATH", "E_SOURCE_BUSY"}:
return CapabilityReport(self.key, FORMAT_ID, "unsafe", root=root)
return None
# Exact absolute path / UUID must not depend on project-dir enumeration
# (issue #19: 2k+ siblings or project buckets would fail probe first).
try:
exact_path = _exact_path_candidate(root, query)
except DiagnosticError as error:
if error.code in {"E_UNSAFE_PATH", "E_SOURCE_BUSY"}:
return CapabilityReport(self.key, FORMAT_ID, "unsafe", root=root)
if (
error.code == "E_NO_MATCH"
and query.ref
and os.path.isabs(query.ref.strip())
):
# Missing approved session path: capability remains supported
# so list/show can surface E_NO_MATCH without sibling scandir.
projects = os.path.join(root, "projects")
if _regular_directory(projects, root):
return CapabilityReport(
self.key, FORMAT_ID, "supported", root=root, evidence=(FORMAT_ID,)
)
return CapabilityReport(self.key, FORMAT_ID, "unavailable", root=root)
exact_path = None
if exact_path is not None:
report = _probe_path(exact_path)
if report is not None:
return report
exact = _exact_uuid_ref(query.ref)
prefer = _prefer_slugs_for(query)
if exact is not None:
# Exact UUID: basename probe only (never scandir session siblings).
if prefer:
for path in _direct_uuid_under_slugs(root, exact, prefer):
report = _probe_path(path)
if report is not None:
return report
for path in _exact_uuid_paths(root, exact, prefer_slugs=prefer):
report = _probe_path(path)
if report is not None:
return report
projects = os.path.join(root, "projects")
if _regular_directory(projects, root):
return CapabilityReport(
self.key, FORMAT_ID, "supported", root=root, evidence=(FORMAT_ID,)
)
return CapabilityReport(self.key, FORMAT_ID, "unavailable", root=root)
paths = _session_paths(root, prefer_slugs=prefer, cwd_scoped=bool(prefer))
if not paths:
# Empty cwd-scoped dir is still a supported store layout if projects exists.
projects = os.path.join(root, "projects")
if _regular_directory(projects, root):
return CapabilityReport(self.key, FORMAT_ID, "supported", root=root, evidence=(FORMAT_ID,))
return CapabilityReport(self.key, FORMAT_ID, "unavailable", root=root)
for path in paths:
report = _probe_path(path)
if report is not None:
return report
return CapabilityReport(self.key, FORMAT_ID, "supported", root=root, evidence=(FORMAT_ID,))
except DiagnosticError as error:
state = "unsafe" if error.code in {"E_UNSAFE_PATH", "E_SOURCE_BUSY"} else "unsupported"
return CapabilityReport(self.key, FORMAT_ID, state)
def list(self, query: Query, budget: ReadBudget) -> list[SessionSummary]:
root = _existing_root(query)
if root is None:
raise DiagnosticError("E_CAPABILITY_UNAVAILABLE", source=self.key, provider=FORMAT_ID)
values: list[SessionSummary] = []
cwd_fallback = False
# Absolute approved path: validate/read that file only (#19).
exact_path = _exact_path_candidate(root, query)
if exact_path is not None:
paths: list[str] = [exact_path]
else:
exact = _exact_uuid_ref(query.ref)
prefer = _prefer_slugs_for(query)
if exact is not None and prefer:
# Fast path: try preferred slug file(s) first without broad
# projects scandir. Only enumerate other buckets when none of
# the direct candidates survive recorded-cwd validation.
direct = _direct_uuid_under_slugs(root, exact, prefer)
for path in direct:
item = _summary(path, root, query, budget)
if item is not None:
values.append(item)
if values:
values.sort(
key=lambda item: (
item.updated_at is None,
item.updated_at or "",
item.session_id,
),
reverse=True,
)
return values
# Direct file missing or cwd-mismatched — broad basename probe.
paths = _exact_uuid_paths(root, exact, prefer_slugs=prefer)
# Already tried direct paths above; skip re-summarizing them.
tried = set(direct)
paths = [path for path in paths if path not in tried]
else:
# Non-exact list still cwd-scopes when a preferred slug exists.
if prefer:
paths, cwd_fallback = _cwd_scoped_session_paths(root, prefer)
else:
paths = _session_paths(root, exact_uuid=exact)
for path in paths:
if cwd_fallback and query.cwd is not None:
prefilter_budget = _provisional_metadata_budget(budget)
if not _recorded_cwd_matches(
path,
root,
query.cwd,
prefilter_budget,
):
_commit_metadata_budget(budget, prefilter_budget)
continue
# List still needs recorded cwd/title for collision safety; show does lineage.
item = _summary(path, root, query, budget)
if item is not None:
values.append(item)
values.sort(key=lambda item: (item.updated_at is None, item.updated_at or "", item.session_id), reverse=True)
return values
def show(self, ref: ResolvedRef, query: Query, budget: ReadBudget) -> Session:
root = _existing_root(query)
if root is None:
raise DiagnosticError("E_CAPABILITY_UNAVAILABLE", source=self.key, provider=FORMAT_ID)
path = ref.source_path
if path is None:
prefer = _prefer_slugs_for(query)
# Prefer direct slug file(s), then broad basename probe only if needed
# (cwd-mismatched direct must not hide a relocated eligible copy).
if prefer:
ordered = _direct_uuid_under_slugs(root, ref.session_id, prefer)
else:
ordered = []
if not ordered:
ordered = _session_paths(
root,
prefer_slugs=prefer,
exact_uuid=ref.session_id,
cwd_scoped=False,
)
else:
# Validate direct candidates; fall back to broad only on cwd miss
# (summary is None). Limit/busy/corrupt diagnostics must propagate
# (Codex P2 on #19) — do not swallow as soft no-match.
chosen: str | None = None
for candidate in ordered:
summary = _summary(candidate, root, query, budget)
if summary is not None and summary.session_id == ref.session_id:
chosen = candidate
break
if chosen is not None:
path = chosen
ordered = []
else:
tried = set(ordered)
ordered = [
item
for item in _exact_uuid_paths(
root, ref.session_id, prefer_slugs=prefer
)
if item not in tried
]
if path is None:
if not ordered:
raise DiagnosticError("E_NO_MATCH", source=self.key, provider=FORMAT_ID)
if len(ordered) == 1:
path = ordered[0]
else:
for candidate in ordered:
summary = _summary(candidate, root, query, budget)
if summary is not None and summary.session_id == ref.session_id:
path = candidate
break
if path is None:
raise DiagnosticError("E_NO_MATCH", source=self.key, provider=FORMAT_ID)
_validate_claude_bounds(budget)
baseline = _provisional_metadata_budget(budget)
provisional = _provisional_metadata_budget(baseline)
observation: FileSnapshot | None = None
try:
observation = snapshot_regular_file(
path,
root=root,
bounds=budget.limits,
attempts=budget.limits.snapshot_attempts,
membership_limit=budget.limits.scanned_records,
budget=provisional,
provider=FORMAT_ID,
)
try:
index = _index_snapshot(observation, provisional)
except BaseException:
observation.close()
raise
except DiagnosticError as error:
if error.code != "E_LIMIT_EXCEEDED":
raise
observation, index = _build_tail_graph(path, root, budget)
with observation:
return _assemble_from_index(
observation, index, path, ref, query, budget
)
# Commit full-path charges before assembly so turns-exhaustion hard-fails.
try:
budget.commit_provisional(baseline, provisional)
return _assemble_from_index(observation, index, path, ref, query, budget)
finally:
observation.close()
ADAPTER = ClaudeAdapter()
SHA-256: 4ee8647cdf0ab1e739fffffae2b5979938595bb472303ee1c785dac2b254acfc