cctally 1.77.0 → 1.79.0

package/bin/_lib_codex_conversation.py

@@ -16,6 +16,8 @@ from __future__ import annotations
 import dataclasses
 import hashlib
 import json
+import math
+import re
 from typing import Any, Iterable
 
 # Reuse the Claude display helpers by IMPORT (never move them). _strip_ansi lives
@@ -65,6 +67,13 @@ CODEX_TEXT_CAP = 16000
 # Display title cap. Equal-by-test to the Claude first-prompt titler's _TITLE_MAX.
 CODEX_TITLE_MAX = _TITLE_MAX
 
+# Bumped when the normalized row/item contract changes in a way that requires
+# deterministic replay of retained Codex events.  ``sync_codex_conversations``
+# stores this in conversations.db only after a successful full sync; a missing
+# or older value arms the existing byte-zero rebuild path without a schema
+# migration (the store is wholly re-derivable).
+CODEX_CONVERSATION_CONTRACT_VERSION = "5"
+
 # Structural wrapper prefixes skipped during title selection (§4.3), pinned from
 # the corpus (title-wrapper-window). Prefix-structural, never content heuristics.
 CODEX_TITLE_SKIP_PREFIXES: tuple[str, ...] = (
@@ -72,7 +81,7 @@ CODEX_TITLE_SKIP_PREFIXES: tuple[str, ...] = (
     "<user_instructions>",
 )
 
-_PROSE_KINDS = frozenset({"user", "assistant", "reasoning"})
+_MIRROR_KINDS = frozenset({"user", "assistant", "reasoning", "meta"})
 
 
 def _cap(text: str) -> tuple[str, bool]:
@@ -175,6 +184,11 @@ def _canonical_json(value: Any) -> str:
         return ""
 
 
+def _reject_json_constant(value: str) -> None:
+    """Reject NaN/Infinity tokens so card JSON stays standards-compliant."""
+    raise ValueError(f"non-finite JSON constant: {value}")
+
+
 def _stringify(value: Any) -> str:
     if value is None:
         return ""
@@ -196,6 +210,636 @@ def _join_content_texts(content: Any) -> str:
     return "\n".join(parts)
 
 
+# ── card-ready native tool shaping (#331 Task A) ─────────────────────────────
+
+CODEX_CARD_SCHEMA_VERSION = 1
+_CARD_MAX_COMMANDS = 8
+_CARD_MAX_PARTS = 128
+_CARD_MAX_FILES = 128
+_CARD_MAX_RESULTS = 50
+_CARD_MAX_COLLECTION = 64
+_CARD_HARNESS_PARSE_CAP = 1_000_000
+_AGENT_OPERATIONS = frozenset({
+    "spawn_agent", "wait_agent", "send_message", "list_agents",
+    "followup_task", "interrupt_agent",
+})
+_EXEC_METADATA_KEYS = frozenset({
+    "justification", "login", "max_output_tokens", "prefix_rule",
+    "sandbox_permissions", "shell", "tty", "yield_time_ms",
+})
+_HARNESS_STATUS_RE = re.compile(
+    r"\A(Script completed|Script failed)\nWall time ([^\n]+)\n\nOutput:\Z")
+
+
+class _LiteralError(ValueError):
+    """Closed-parser rejection; callers preserve the raw provider payload."""
+
+
+class _TextBudget:
+    def __init__(self, limit: int):
+        self.remaining = max(0, int(limit))
+        self.truncated = False
+
+    def take(self, value: str) -> str:
+        if len(value) <= self.remaining:
+            self.remaining -= len(value)
+            return value
+        kept = value[:self.remaining]
+        self.remaining = 0
+        self.truncated = True
+        return kept
+
+
+class _HarnessLiteralParser:
+    """Tiny non-executing parser for the JSON-like values in Codex harnesses.
+
+    It accepts only objects, arrays, double-quoted JSON strings, finite JSON
+    numbers, booleans and null.  Object keys may be unquoted harness tokens
+    (including ``yield_time-ms``-style hyphens).  Expressions, identifiers as
+    values, templates, comments and trailing code reject the shape.
+    """
+
+    def __init__(self, source: str, pos: int = 0, *, max_depth: int = 8):
+        self.source = source
+        self.pos = pos
+        self.max_depth = max_depth
+
+    def _ws(self) -> None:
+        size = len(self.source)
+        while self.pos < size and self.source[self.pos].isspace():
+            self.pos += 1
+
+    def _consume(self, token: str) -> None:
+        self._ws()
+        if not self.source.startswith(token, self.pos):
+            raise _LiteralError(token)
+        self.pos += len(token)
+
+    def _string(self) -> str:
+        self._ws()
+        if self.pos >= len(self.source) or self.source[self.pos] != '"':
+            raise _LiteralError("string")
+        try:
+            value, end = json.JSONDecoder().raw_decode(self.source, self.pos)
+        except (json.JSONDecodeError, TypeError) as exc:
+            raise _LiteralError("string") from exc
+        if not isinstance(value, str):
+            raise _LiteralError("string")
+        self.pos = end
+        return value
+
+    def _key(self) -> str:
+        self._ws()
+        if self.pos < len(self.source) and self.source[self.pos] == '"':
+            return self._string()
+        match = re.match(r"[A-Za-z_$][A-Za-z0-9_$-]*", self.source[self.pos:])
+        if match is None:
+            raise _LiteralError("key")
+        self.pos += len(match.group(0))
+        return match.group(0)
+
+    def value(self, depth: int = 0) -> Any:
+        if depth > self.max_depth:
+            raise _LiteralError("depth")
+        self._ws()
+        if self.pos >= len(self.source):
+            raise _LiteralError("value")
+        char = self.source[self.pos]
+        if char == '"':
+            return self._string()
+        if char == "{":
+            self.pos += 1
+            result: dict[str, Any] = {}
+            self._ws()
+            if self.pos < len(self.source) and self.source[self.pos] == "}":
+                self.pos += 1
+                return result
+            while len(result) < 64:
+                key = self._key()
+                if key in result:
+                    raise _LiteralError("duplicate key")
+                self._consume(":")
+                result[key] = self.value(depth + 1)
+                self._ws()
+                if self.pos < len(self.source) and self.source[self.pos] == "}":
+                    self.pos += 1
+                    return result
+                self._consume(",")
+            raise _LiteralError("object size")
+        if char == "[":
+            self.pos += 1
+            result_list: list[Any] = []
+            self._ws()
+            if self.pos < len(self.source) and self.source[self.pos] == "]":
+                self.pos += 1
+                return result_list
+            while len(result_list) < 64:
+                result_list.append(self.value(depth + 1))
+                self._ws()
+                if self.pos < len(self.source) and self.source[self.pos] == "]":
+                    self.pos += 1
+                    return result_list
+                self._consume(",")
+            raise _LiteralError("array size")
+        for token, value in (("true", True), ("false", False), ("null", None)):
+            if self.source.startswith(token, self.pos):
+                self.pos += len(token)
+                return value
+        match = re.match(r"-?(?:0|[1-9][0-9]*)(?:\.[0-9]+)?(?:[eE][+-]?[0-9]+)?",
+                         self.source[self.pos:])
+        if match is None:
+            raise _LiteralError("literal")
+        token = match.group(0)
+        if len(token) > 256:
+            raise _LiteralError("number length")
+        self.pos += len(token)
+        try:
+            value = float(token) if any(c in token for c in ".eE") else int(token)
+        except (OverflowError, ValueError) as exc:
+            raise _LiteralError("number") from exc
+        if isinstance(value, float) and not (-float("inf") < value < float("inf")):
+            raise _LiteralError("finite number")
+        return value
+
+
+def _bounded_metadata(value: Any, budget: _TextBudget) -> Any:
+    if isinstance(value, str):
+        return budget.take(value)
+    if isinstance(value, (bool, int, float)):
+        return value
+    if (isinstance(value, list) and len(value) <= 32
+            and all(isinstance(part, str) for part in value)):
+        return [budget.take(part) for part in value]
+    return None
+
+
+def _bounded_json(value: Any, budget: _TextBudget, *, depth: int = 0) -> Any:
+    """Retain JSON structure under one shared text/shape budget.
+
+    Native payload readback remains authoritative; this projection is only the
+    bounded card wire. Collection overflow is explicit through the caller's
+    ``budget.truncated`` flag rather than silently expanding the detail route.
+    """
+    if depth > 8:
+        budget.truncated = True
+        return None
+    if isinstance(value, str):
+        return budget.take(value)
+    if value is None or isinstance(value, (bool, int)):
+        return value
+    if isinstance(value, float):
+        if math.isfinite(value):
+            return value
+        budget.truncated = True
+        return None
+    if isinstance(value, list):
+        if len(value) > _CARD_MAX_COLLECTION:
+            budget.truncated = True
+        return [
+            _bounded_json(part, budget, depth=depth + 1)
+            for part in value[:_CARD_MAX_COLLECTION]
+        ]
+    if isinstance(value, dict):
+        if len(value) > _CARD_MAX_COLLECTION:
+            budget.truncated = True
+        result = {}
+        for key, part in list(value.items())[:_CARD_MAX_COLLECTION]:
+            if not isinstance(key, str):
+                budget.truncated = True
+                continue
+            result[budget.take(key)] = _bounded_json(
+                part, budget, depth=depth + 1)
+        return result
+    budget.truncated = True
+    return None
+
+
+def _json_object(value: Any) -> dict | None:
+    if isinstance(value, dict):
+        return value
+    if not isinstance(value, str) or len(value) > _CARD_HARNESS_PARSE_CAP:
+        return None
+    try:
+        parsed = json.loads(value, parse_constant=_reject_json_constant)
+    except (json.JSONDecodeError, TypeError, ValueError):
+        return None
+    return parsed if isinstance(parsed, dict) else None
+
+
+def decode_secondary_tool_call_card(
+    payload: dict, *, text_cap: int = CODEX_TEXT_CAP,
+) -> dict | None:
+    """Bounded additive wire for plans, web actions, and agent operations.
+
+    Unknown or malformed shapes deliberately return ``None`` so the existing
+    provider-name + raw-argument fallback remains visible and payload readback
+    stays authoritative.
+    """
+    if not isinstance(payload, dict):
+        return None
+    ptype = payload.get("type")
+    name = payload.get("name")
+    status = payload.get("status") if isinstance(payload.get("status"), str) else "requested"
+    budget = _TextBudget(text_cap)
+    if ptype == "function_call" and name == "update_plan":
+        arguments = _json_object(payload.get("arguments"))
+        plan = arguments.get("plan") if isinstance(arguments, dict) else None
+        if not isinstance(plan, list) or len(plan) > _CARD_MAX_COLLECTION:
+            return None
+        items = []
+        for item in plan:
+            if not (isinstance(item, dict)
+                    and isinstance(item.get("step"), str)
+                    and isinstance(item.get("status"), str)):
+                return None
+            items.append({
+                "step": budget.take(item["step"]),
+                "status": budget.take(item["status"]),
+            })
+        card = {
+            "schema_version": CODEX_CARD_SCHEMA_VERSION,
+            "type": "plan", "source": "update_plan",
+            "call_status": status, "items": items,
+        }
+        explanation = arguments.get("explanation")
+        if explanation is not None:
+            if not isinstance(explanation, str):
+                return None
+            card["explanation"] = budget.take(explanation)
+        if budget.truncated:
+            card["truncated"] = True
+        return card
+    if ptype == "web_search_call":
+        action = payload.get("action")
+        if not isinstance(action, (dict, str)):
+            return None
+        bounded_action = _bounded_json(action, budget)
+        card = {
+            "schema_version": CODEX_CARD_SCHEMA_VERSION,
+            "type": "web_search", "source": "web_search_call",
+            "call_status": status, "action": bounded_action,
+        }
+        if isinstance(action, dict) and isinstance(action.get("query"), str):
+            card["query"] = budget.take(action["query"])
+        if budget.truncated:
+            card["truncated"] = True
+        return card
+    if ptype == "function_call" and name in _AGENT_OPERATIONS:
+        arguments = _json_object(payload.get("arguments"))
+        if arguments is None:
+            return None
+        bounded = _bounded_json(arguments, budget)
+        card = {
+            "schema_version": CODEX_CARD_SCHEMA_VERSION,
+            "type": "agent", "operation": name,
+            "call_status": status, "arguments": bounded,
+        }
+        if budget.truncated:
+            card["truncated"] = True
+        return card
+    return None
+
+
+def decode_secondary_tool_result(
+    payload: dict, *, text_cap: int = CODEX_TEXT_CAP,
+) -> dict | None:
+    if not isinstance(payload, dict) or payload.get("type") not in _RESPONSE_TOOL_OUTPUTS:
+        return None
+    value = payload["output"] if "output" in payload else payload.get("tools")
+    if isinstance(value, str) and len(value) <= _CARD_HARNESS_PARSE_CAP:
+        try:
+            parsed = json.loads(value, parse_constant=_reject_json_constant)
+        except (json.JSONDecodeError, TypeError, ValueError):
+            parsed = value
+    else:
+        parsed = value
+    budget = _TextBudget(text_cap)
+    bounded = _bounded_json(parsed, budget)
+    status = payload.get("status") if isinstance(payload.get("status"), str) else "returned"
+    return {"status": status, "value": bounded, "truncated": budget.truncated}
+
+
+def decode_secondary_event_card(
+    payload: dict, *, text_cap: int = CODEX_TEXT_CAP,
+) -> dict | None:
+    if not isinstance(payload, dict):
+        return None
+    ptype = payload.get("type")
+    budget = _TextBudget(text_cap)
+    if ptype == "web_search_end":
+        action = payload.get("action")
+        query = payload.get("query")
+        results = payload.get("results", [])
+        if not isinstance(action, (dict, str)) or not isinstance(query, str):
+            return None
+        if not isinstance(results, list):
+            return None
+        if len(results) > _CARD_MAX_RESULTS:
+            budget.truncated = True
+        card = {
+            "schema_version": CODEX_CARD_SCHEMA_VERSION,
+            "type": "web_search_completion", "source": "web_search_end",
+            "status": (payload.get("status")
+                       if isinstance(payload.get("status"), str) else "returned"),
+            "query": budget.take(query),
+            "action": _bounded_json(action, budget),
+            "results": [
+                _bounded_json(result, budget)
+                for result in results[:_CARD_MAX_RESULTS]
+            ],
+        }
+        if "error" in payload:
+            card["error"] = _bounded_json(payload.get("error"), budget)
+        if budget.truncated:
+            card["truncated"] = True
+        return card
+    if ptype == "mcp_tool_call_end":
+        invocation = payload.get("invocation")
+        if not isinstance(invocation, dict):
+            return None
+        server = invocation.get("server")
+        tool = invocation.get("tool") or invocation.get("name")
+        if server is not None and not isinstance(server, str):
+            return None
+        if not isinstance(tool, str):
+            return None
+        result = payload.get("result")
+        if isinstance(result, dict) and "Ok" in result:
+            status = "ok"
+        elif isinstance(result, dict) and any(key in result for key in ("Err", "Error")):
+            status = "error"
+        else:
+            status = (payload.get("status")
+                      if isinstance(payload.get("status"), str) else "returned")
+        card = {
+            "schema_version": CODEX_CARD_SCHEMA_VERSION,
+            "type": "mcp_completion", "source": "mcp_tool_call_end",
+            "status": status,
+            "server": budget.take(server) if isinstance(server, str) else None,
+            "tool": budget.take(tool),
+            "arguments": _bounded_json(invocation.get("arguments"), budget),
+            "result": _bounded_json(result, budget),
+            "duration": _bounded_json(payload.get("duration"), budget),
+        }
+        if budget.truncated:
+            card["truncated"] = True
+        return card
+    return None
+
+
+def _exec_invocations(source: str, *, budget: _TextBudget) -> list[dict] | None:
+    """Decode only the complete current exec harness statement grammar."""
+    commands: list[dict] = []
+    pos = 0
+    while True:
+        prefix = re.match(
+            r"\s*const\s+([A-Za-z_$][A-Za-z0-9_$]*)\s*=\s*"
+            r"await\s+tools\.exec_command",
+            source[pos:],
+        )
+        if prefix is None or len(commands) >= _CARD_MAX_COMMANDS:
+            return None
+        variable = prefix.group(1)
+        parser = _HarnessLiteralParser(source, pos + prefix.end())
+        try:
+            parser._consume("(")
+            value = parser.value()
+            parser._consume(")")
+        except _LiteralError:
+            return None
+        if not isinstance(value, dict) or not isinstance(value.get("cmd"), str):
+            return None
+        workdir = value.get("workdir")
+        if workdir is not None and not isinstance(workdir, str):
+            return None
+        command = {
+            "workdir": budget.take(workdir) if isinstance(workdir, str) else None,
+            "command": "",
+            "metadata": {},
+        }
+        command["command"] = budget.take(value["cmd"])
+        for key in sorted(_EXEC_METADATA_KEYS):
+            if key not in value:
+                continue
+            bounded = _bounded_metadata(value[key], budget)
+            if bounded is not None:
+                command["metadata"][key] = bounded
+        commands.append(command)
+        suffix = re.match(
+            r"\s*;\s*text\s*\(\s*" + re.escape(variable)
+            + r"\.output\s*\)\s*;?",
+            source[parser.pos:],
+        )
+        if suffix is None:
+            return None
+        pos = parser.pos + suffix.end()
+        if not source[pos:].strip():
+            return commands
+
+
+def _decode_apply_patch_program(source: str) -> str | None:
+    """Recognize the exact current ``const patch`` apply_patch harness."""
+    match = re.match(r"\A\s*const\s+([A-Za-z_$][A-Za-z0-9_$]*)\s*=\s*", source)
+    if match is None:
+        return None
+    parser = _HarnessLiteralParser(source, match.end())
+    try:
+        patch = parser._string()
+    except _LiteralError:
+        return None
+    variable = match.group(1)
+    tail = source[parser.pos:]
+    pattern = (
+        r"\s*;\s*text\s*\(\s*await\s+tools\.apply_patch\s*\(\s*"
+        + re.escape(variable)
+        + r"\s*\)\s*\)\s*;?\s*\Z"
+    )
+    return patch if re.fullmatch(pattern, tail) else None
+
+
+def _apply_patch_heredoc(command: str) -> str | None:
+    for pattern in (
+        r"\A\s*apply_patch\s+<<'([A-Za-z_][A-Za-z0-9_]*)'\n(.*)\n\1\s*\Z",
+        r'\A\s*apply_patch\s+<<"([A-Za-z_][A-Za-z0-9_]*)"\n(.*)\n\1\s*\Z',
+        r"\A\s*apply_patch\s+<<([A-Za-z_][A-Za-z0-9_]*)\n(.*)\n\1\s*\Z",
+    ):
+        match = re.fullmatch(pattern, command, re.DOTALL)
+        if match is not None:
+            return match.group(2)
+    return None
+
+
+def _patch_files_from_apply_patch(
+    patch: str, budget: _TextBudget | None = None,
+) -> list[dict]:
+    files: list[dict] = []
+    for line in patch.splitlines():
+        match = re.match(r"\*\*\* (Add|Update|Delete) File: (.+)\Z", line)
+        if match is not None and len(files) < _CARD_MAX_FILES:
+            status = {"Add": "added", "Update": "modified", "Delete": "deleted"}[
+                match.group(1)]
+            path = budget.take(match.group(2)) if budget is not None else match.group(2)
+            files.append({"path": path, "status": status})
+            continue
+        move = re.match(r"\*\*\* Move to: (.+)\Z", line)
+        if move is not None and files:
+            files[-1]["move_path"] = (
+                budget.take(move.group(1)) if budget is not None else move.group(1))
+            files[-1]["status"] = "moved"
+    return files
+
+
+def _complete_apply_patch(patch: str) -> bool:
+    """Require one closed apply_patch envelope with at least one file action."""
+    lines = patch.splitlines()
+    if (not lines or lines[0] != "*** Begin Patch"
+            or lines[-1] != "*** End Patch"):
+        return False
+    if lines.count("*** Begin Patch") != 1 or lines.count("*** End Patch") != 1:
+        return False
+    return any(re.fullmatch(r"\*\*\* (?:Add|Update|Delete) File: .+", line)
+               for line in lines[1:-1])
+
+
+def decode_tool_call_card(payload: dict, *, text_cap: int = CODEX_TEXT_CAP) -> dict | None:
+    """Return the additive card contract for a structurally proven call."""
+    if not isinstance(payload, dict) or payload.get("type") != "custom_tool_call":
+        return None
+    name = payload.get("name")
+    value = payload.get("input")
+    if not isinstance(name, str) or not isinstance(value, str):
+        return None
+    if len(value) > _CARD_HARNESS_PARSE_CAP:
+        return None
+    budget = _TextBudget(text_cap)
+    status = payload.get("status") if isinstance(payload.get("status"), str) else "unknown"
+    if name == "apply_patch" and _complete_apply_patch(value):
+        files = _patch_files_from_apply_patch(value, budget)
+        patch = budget.take(value)
+        return {
+            "schema_version": CODEX_CARD_SCHEMA_VERSION,
+            "type": "patch", "source": "apply_patch", "status": status,
+            "patch": patch, "files": files,
+            "truncated": budget.truncated,
+        }
+    if name != "exec":
+        return None
+    patch_value = _decode_apply_patch_program(value)
+    if patch_value is not None and _complete_apply_patch(patch_value):
+        files = _patch_files_from_apply_patch(patch_value, budget)
+        patch = budget.take(patch_value)
+        return {
+            "schema_version": CODEX_CARD_SCHEMA_VERSION,
+            "type": "patch", "source": "tools.apply_patch", "status": status,
+            "patch": patch, "files": files,
+            "truncated": budget.truncated,
+        }
+    commands = _exec_invocations(value, budget=budget)
+    if commands is None:
+        return None
+    if len(commands) == 1:
+        heredoc = _apply_patch_heredoc(commands[0]["command"])
+        if heredoc is not None and _complete_apply_patch(heredoc):
+            files = _patch_files_from_apply_patch(heredoc, budget)
+            return {
+                "schema_version": CODEX_CARD_SCHEMA_VERSION,
+                "type": "patch", "source": "exec_apply_patch", "status": status,
+                "patch": budget.take(heredoc), "files": files,
+                "workdir": commands[0]["workdir"], "truncated": budget.truncated,
+            }
+    card = {
+        "schema_version": CODEX_CARD_SCHEMA_VERSION,
+        "type": "terminal", "status": status, "commands": commands,
+    }
+    if budget.truncated:
+        card["truncated"] = True
+    return card
+
+
+def decode_tool_output_card(
+    payload: dict, *, text_cap: int = CODEX_TEXT_CAP,
+) -> tuple[dict, str] | None:
+    """Unwrap supported output envelopes without losing malformed parts."""
+    if not isinstance(payload, dict) or payload.get("type") not in _RESPONSE_TOOL_OUTPUTS:
+        return None
+    value = payload["output"] if "output" in payload else payload.get("tools")
+    values = value if isinstance(value, list) else [value]
+    budget = _TextBudget(text_cap)
+    parts: list[dict] = []
+    status = payload.get("status") if isinstance(payload.get("status"), str) else "unknown"
+    for index, part in enumerate(values[:_CARD_MAX_PARTS]):
+        if isinstance(part, str):
+            text = part
+            match = _HARNESS_STATUS_RE.fullmatch(text) if index == 0 else None
+            if match is not None:
+                status = "completed" if match.group(1) == "Script completed" else "failed"
+                continue
+            parts.append({"type": "text", "stream": "output", "text": budget.take(text)})
+            continue
+        if isinstance(part, dict) and part.get("type") == "input_text" \
+                and isinstance(part.get("text"), str):
+            text = part["text"]
+            match = _HARNESS_STATUS_RE.fullmatch(text) if index == 0 else None
+            if match is not None:
+                status = "completed" if match.group(1) == "Script completed" else "failed"
+                continue
+            stream = part.get("stream") if part.get("stream") in {"stdout", "stderr"} else "output"
+            parts.append({"type": "text", "stream": stream, "text": budget.take(text)})
+            continue
+        raw = _canonical_json(part)
+        parts.append({"type": "raw", "stream": "output", "text": budget.take(raw)})
+    if len(values) > _CARD_MAX_PARTS:
+        budget.truncated = True
+    card = {
+        "schema_version": CODEX_CARD_SCHEMA_VERSION,
+        "type": "terminal_output", "status": status,
+        "is_error": status in {"failed", "error"},
+        "parts": parts, "truncated": budget.truncated,
+    }
+    return card, "".join(part["text"] for part in parts)
+
+
+def decode_patch_event_card(payload: dict, *, text_cap: int = CODEX_TEXT_CAP) -> dict | None:
+    """Bounded, provider-truthful ``patch_apply_end`` projection."""
+    if not isinstance(payload, dict) or payload.get("type") != "patch_apply_end":
+        return None
+    budget = _TextBudget(text_cap)
+    files: list[dict] = []
+    changes = payload.get("changes")
+    if isinstance(changes, list):
+        for change in changes[:_CARD_MAX_FILES]:
+            if not isinstance(change, dict):
+                files.append({"raw": budget.take(_canonical_json(change))})
+                continue
+            entry: dict[str, Any] = {}
+            for key in ("path", "move_path", "status"):
+                if isinstance(change.get(key), str):
+                    entry[key] = budget.take(change[key])
+            if isinstance(change.get("unified_diff"), str):
+                entry["unified_diff"] = budget.take(change["unified_diff"])
+            unknown = {key: value for key, value in change.items()
+                       if key not in {"path", "move_path", "status", "unified_diff"}}
+            if unknown:
+                entry["raw_extra"] = budget.take(_canonical_json(unknown))
+            files.append(entry)
+        if len(changes) > _CARD_MAX_FILES:
+            budget.truncated = True
+    elif "changes" in payload:
+        files.append({"raw": budget.take(_canonical_json(changes))})
+    stdout = budget.take(payload["stdout"]) if isinstance(payload.get("stdout"), str) else None
+    stderr = budget.take(payload["stderr"]) if isinstance(payload.get("stderr"), str) else None
+    status = payload.get("status") if isinstance(payload.get("status"), str) else "unknown"
+    success = payload.get("success") if isinstance(payload.get("success"), bool) else None
+    return {
+        "schema_version": CODEX_CARD_SCHEMA_VERSION,
+        "type": "patch", "source": "patch_apply_end",
+        "files": files,
+        "has_diff": any(isinstance(entry.get("unified_diff"), str) for entry in files),
+        "status": status, "success": success, "stdout": stdout, "stderr": stderr,
+        "truncated": budget.truncated,
+    }
+
+
 @dataclasses.dataclass
 class _Extracted:
     kind: str
@@ -203,6 +847,334 @@ class _Extracted:
     column: str                # "text" | "search_tool" | "search_thinking"
     detail: dict | None
     touches: list[tuple[str, str]]  # (file_path, tool)
+    identity_text: str | None = None  # raw pre-segmentation mirror/payload identity
+
+
+_REASONING_TITLE_RE = re.compile(r"\A\*\*([^\n]+)\*\*\Z")
+_MARKER_DIRECTIVE_RE = re.compile(r"\A::([a-z0-9-]+)\{(.*)\}\Z")
+_MARKER_ATTR_RE = re.compile(
+    r'([A-Za-z][A-Za-z0-9]*)=("(?:[^"\\]|\\.)*"|true|false)')
+_MARKER_DIRECTIVES = {
+    "git-create-branch": ({"cwd", "branch"}, "create_branch"),
+    "git-stage": ({"cwd"}, "stage"),
+    "git-commit": ({"cwd"}, "commit"),
+    "git-push": ({"cwd", "branch"}, "push"),
+    "git-create-pr": ({"cwd", "branch", "url", "isDraft"}, "create_pr"),
+}
+_MEM_CITATION_RE = re.compile(
+    r"\A[^<>\r\n]+:\d+-\d+\|note=\[[^\]\r\n]*\]\Z")
+_ROLLOUT_ID_RE = re.compile(
+    r"\A[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}\Z",
+    re.IGNORECASE,
+)
+
+
+def _reasoning_projection(
+    *, source: str, summary: str = "", body: str = "",
+) -> dict | None:
+    """Provider-truthful title/summary/body projection for non-empty reasoning."""
+    summary_nonblank = summary.strip()
+    body_nonblank = body.strip()
+    if not summary_nonblank and not body_nonblank:
+        return None
+    result = {"schema_version": 1, "source": source}
+    match = _REASONING_TITLE_RE.fullmatch(summary_nonblank)
+    if match is not None:
+        title = match.group(1)
+        if title.strip() == title and "**" not in title:
+            result["title"] = title
+        else:
+            result["summary"] = summary_nonblank
+    elif summary_nonblank:
+        result["summary"] = summary_nonblank
+    if body_nonblank:
+        result["body"] = body_nonblank
+    return result
+
+
+def _parse_marker_directive(line: str) -> dict | None:
+    matched = _MARKER_DIRECTIVE_RE.fullmatch(line)
+    if matched is None or matched.group(1) not in _MARKER_DIRECTIVES:
+        return None
+    name, source = matched.group(1), matched.group(2)
+    attrs: dict[str, object] = {}
+    pos = 0
+    while pos < len(source):
+        while pos < len(source) and source[pos] == " ":
+            pos += 1
+        token = _MARKER_ATTR_RE.match(source, pos)
+        if token is None or token.group(1) in attrs:
+            return None
+        raw = token.group(2)
+        try:
+            value = json.loads(raw) if raw.startswith('"') else raw == "true"
+        except (json.JSONDecodeError, TypeError):
+            return None
+        attrs[token.group(1)] = value
+        pos = token.end()
+        if pos < len(source) and source[pos] != " ":
+            return None
+    required, action = _MARKER_DIRECTIVES[name]
+    if set(attrs) != required:
+        return None
+    if any(not isinstance(attrs[key], str) or not attrs[key]
+           for key in required - {"isDraft"}):
+        return None
+    if name == "git-create-pr":
+        if not isinstance(attrs.get("isDraft"), bool):
+            return None
+        if not str(attrs.get("url", "")).startswith(("https://", "http://")):
+            return None
+    marker = {"schema_version": 1, "type": "git", "action": action}
+    if name == "git-create-pr":
+        marker["draft"] = attrs["isDraft"]
+    return marker
+
+
+def _parse_memory_citation(lines: list[str]) -> dict | None:
+    try:
+        citation_open = lines.index("<citation_entries>")
+        citation_close = lines.index("</citation_entries>")
+        rollout_open = lines.index("<rollout_ids>")
+        rollout_close = lines.index("</rollout_ids>")
+    except ValueError:
+        return None
+    if not (
+        lines[0] == "<oai-mem-citation>"
+        and lines[-1] == "</oai-mem-citation>"
+        and citation_open == 1
+        and citation_open < citation_close < rollout_open < rollout_close
+        and rollout_close == len(lines) - 2
+        and rollout_open == citation_close + 1
+    ):
+        return None
+    citations = lines[citation_open + 1:citation_close]
+    rollouts = lines[rollout_open + 1:rollout_close]
+    if not citations or not all(_MEM_CITATION_RE.fullmatch(line) for line in citations):
+        return None
+    if not all(_ROLLOUT_ID_RE.fullmatch(line) for line in rollouts):
+        return None
+    return {
+        "schema_version": 1, "type": "memory_citation",
+        "citation_count": len(citations), "rollout_count": len(rollouts),
+    }
+
+
+def _segment_harness_markers(text: str) -> tuple[str, list[dict]]:
+    """Segment only a closed trailing suffix; authored/fenced lookalikes stay text."""
+    lines = text.splitlines()
+    end = len(lines)
+    while end and not lines[end - 1].strip():
+        end -= 1
+    reversed_markers: list[dict] = []
+    while end:
+        marker = _parse_marker_directive(lines[end - 1])
+        start = end - 1
+        if marker is None and lines[end - 1] == "</oai-mem-citation>":
+            starts = [index for index in range(end - 2, -1, -1)
+                      if lines[index] == "<oai-mem-citation>"]
+            if starts:
+                start = starts[0]
+                marker = _parse_memory_citation(lines[start:end])
+        if marker is None:
+            break
+        # A suffix that begins inside an open Markdown fence is authored code.
+        fence: str | None = None
+        for line in lines[:start]:
+            opened = re.match(r"\s*(`{3,}|~{3,})", line)
+            if opened is None:
+                continue
+            char = opened.group(1)[0]
+            if fence is None:
+                fence = char
+            elif fence == char:
+                fence = None
+        if fence is not None:
+            break
+        reversed_markers.append(marker)
+        end = start
+        while end and not lines[end - 1].strip():
+            end -= 1
+    if not reversed_markers:
+        return text, []
+    return "\n".join(lines[:end]).rstrip(), list(reversed(reversed_markers))
+
+
+_TASK_STARTED_KEYS = frozenset({
+    "type", "collaboration_mode_kind", "model_context_window", "started_at", "turn_id",
+})
+_TASK_COMPLETE_KEYS = frozenset({
+    "type", "completed_at", "duration_ms", "last_agent_message", "started_at",
+    "time_to_first_token_ms", "turn_id",
+})
+
+
+def _lifecycle_projection(payload: dict) -> dict:
+    """Safe lifecycle detail plus internal evidence used by canonical folding."""
+    event = payload.get("type")
+    public: dict[str, object] = {
+        "schema_version": 1, "event": event,
+        "state": "started" if event == "task_started" else "completed",
+    }
+    allowed = _TASK_STARTED_KEYS if event == "task_started" else _TASK_COMPLETE_KEYS
+    valid = set(payload) <= allowed
+    turn_id = payload.get("turn_id")
+    valid = valid and isinstance(turn_id, str) and bool(turn_id)
+    if event == "task_started":
+        for src, dst in (
+            ("started_at", "at"),
+            ("collaboration_mode_kind", "collaboration_mode_kind"),
+        ):
+            value = payload.get(src)
+            if value is not None:
+                if src == "started_at" and (
+                    isinstance(value, (int, float)) and not isinstance(value, bool)
+                    and math.isfinite(value)
+                ):
+                    public[dst] = value
+                elif not isinstance(value, str):
+                    valid = False
+                else:
+                    public[dst] = value
+        window = payload.get("model_context_window")
+        if window is not None:
+            if not isinstance(window, int) or isinstance(window, bool):
+                valid = False
+            else:
+                public["model_context_window"] = window
+        return {"lifecycle": public, "_lifecycle_foldable": valid}
+
+    for src, dst in (("completed_at", "at"), ("started_at", "started_at")):
+        value = payload.get(src)
+        if value is not None:
+            if (isinstance(value, (int, float)) and not isinstance(value, bool)
+                    and math.isfinite(value)):
+                public[dst] = value
+            elif not isinstance(value, str):
+                valid = False
+            else:
+                public[dst] = value
+    for key in ("duration_ms", "time_to_first_token_ms"):
+        value = payload.get(key)
+        if value is not None:
+            if (not isinstance(value, (int, float)) or isinstance(value, bool)
+                    or not math.isfinite(value)):
+                valid = False
+            else:
+                public[key] = value
+    message = payload.get("last_agent_message")
+    if message is not None and not isinstance(message, str):
+        valid = False
+        message = ""
+    message = message or ""
+    if message.strip():
+        public["message"] = message[:CODEX_TEXT_CAP]
+    error = payload.get("error")
+    if error is not None:
+        valid = False
+        public["state"] = "failed"
+        public["error"] = _stringify(error)[:CODEX_TEXT_CAP]
+    return {
+        "lifecycle": public,
+        "_lifecycle_foldable": valid,
+        "_lifecycle_message_digest": content_digest(message),
+        "_lifecycle_message_len": content_len(message),
+    }
+
+
+_SKILL_NAME_RE = re.compile(r"<name>\s*([^<\n]+?)\s*</name>", re.IGNORECASE)
+_AGENTS_ENVELOPE_RE = re.compile(
+    r"\A\s*# AGENTS\.md instructions for [^\n]+\n\s*"
+    r"<INSTRUCTIONS>.*?</INSTRUCTIONS>\s*\Z",
+    re.DOTALL,
+)
+_AGENTS_ENV_BUNDLE_RE = re.compile(
+    r"\A\s*# AGENTS\.md instructions for [^\n]+\n\s*"
+    r"<INSTRUCTIONS>.*?</INSTRUCTIONS>\s*"
+    r"<environment_context>.*?</environment_context>\s*\Z",
+    re.DOTALL,
+)
+_CONTEXT_BUNDLE_RE = re.compile(
+    r"\A\s*"
+    r"<recommended_plugins>.*?</recommended_plugins>\s*"
+    r"# AGENTS\.md instructions for [^\n]+\n\s*"
+    r"<INSTRUCTIONS>.*?</INSTRUCTIONS>\s*"
+    r"<environment_context>.*?</environment_context>"
+    r"\s*\Z",
+    re.DOTALL,
+)
+
+
+def _exact_wrapper(text: str, opening: str, closing: str) -> bool:
+    stripped = (text or "").strip()
+    return stripped.startswith(opening) and stripped.endswith(closing)
+
+
+def _meta_detail(meta_kind: str, meta_label: str, text: str) -> dict:
+    detail = {"meta_kind": meta_kind, "meta_label": meta_label}
+    if meta_label == "skill":
+        match = _SKILL_NAME_RE.search(text or "")
+        if match:
+            detail["skill_name"] = match.group(1).strip()
+    return detail
+
+
+def _classify_injected_message(role: str | None, text: str) -> dict | None:
+    """Return an explicit neutral meta descriptor for proven harness content.
+
+    Non-human roles are authoritative provider metadata and always context.  A
+    user/assistant row is reclassified only by a closed, structural wrapper
+    shape; unknown XML and ordinary prose remain provider-authored content.
+    """
+    stripped = (text or "").strip()
+    lower = stripped.lower()
+
+    # Some Codex hosts combine three independently injected envelopes into one
+    # user-role message.  Match the complete, ordered grammar so an appended
+    # human prompt (or merely XML-looking prose) cannot be hidden as context.
+    if role == "user" and _CONTEXT_BUNDLE_RE.fullmatch(text or ""):
+        detail = _meta_detail("context", "context_bundle", text)
+        detail["meta_sections"] = ["plugins", "agents", "environment"]
+        return detail
+    if role == "user" and _AGENTS_ENV_BUNDLE_RE.fullmatch(text or ""):
+        detail = _meta_detail("context", "context_bundle", text)
+        detail["meta_sections"] = ["agents", "environment"]
+        return detail
+
+    if role not in (None, "user", "assistant"):
+        if _exact_wrapper(lower, "<permissions instructions>",
+                          "</permissions instructions>"):
+            return _meta_detail("context", "permissions", text)
+        if _exact_wrapper(lower, "<multi_agent_mode>", "</multi_agent_mode>"):
+            return _meta_detail("context", "mode", text)
+        if _exact_wrapper(lower, "<codex_delegation>", "</codex_delegation>"):
+            return _meta_detail("context", "delegation", text)
+        if _exact_wrapper(lower, "<heartbeat>", "</heartbeat>"):
+            return _meta_detail("notification", "heartbeat", text)
+        if _exact_wrapper(lower, "<model_switch>", "</model_switch>"):
+            return _meta_detail("notification", "model_switch", text)
+        return _meta_detail("context", "role", text)
+
+    wrappers = (
+        ("<environment_context>", "</environment_context>", "context", "environment"),
+        ("<user_instructions>", "</user_instructions>", "context", "instructions"),
+        ("<permissions instructions>", "</permissions instructions>", "context", "permissions"),
+        ("<multi_agent_mode>", "</multi_agent_mode>", "context", "mode"),
+        ("<codex_delegation>", "</codex_delegation>", "context", "delegation"),
+        ("<heartbeat>", "</heartbeat>", "notification", "heartbeat"),
+        ("<model_switch>", "</model_switch>", "notification", "model_switch"),
+        ("<recommended_plugins>", "</recommended_plugins>", "context", "plugins"),
+        ("<skill>", "</skill>", "skill", "skill"),
+        ("<oai-mem-citation>", "</oai-mem-citation>", "notification", "memory"),
+        ("<memory_context>", "</memory_context>", "context", "memory"),
+        ("<memory-context>", "</memory-context>", "context", "memory"),
+    )
+    for opening, closing, meta_kind, label in wrappers:
+        if _exact_wrapper(lower, opening, closing):
+            return _meta_detail(meta_kind, label, text)
+    if role == "user" and _AGENTS_ENVELOPE_RE.fullmatch(text or ""):
+        return _meta_detail("context", "agents", text)
+    return None
 
 
 def _extract(record_type: str | None, payload: dict) -> _Extracted | None:
@@ -213,14 +1185,28 @@ def _extract(record_type: str | None, payload: dict) -> _Extracted | None:
     if record_type == "response_item":
         if ptype == "message":
             role = payload.get("role")
-            kind = "user" if role == "user" else "assistant"
             text = _join_content_texts(payload.get("content"))
+            meta = _classify_injected_message(
+                role if isinstance(role, str) else None, text)
+            if meta is not None:
+                return _Extracted("meta", text, "text", meta, [])
+            kind = "user" if role == "user" else "assistant"
+            if kind == "assistant":
+                clean, markers = _segment_harness_markers(text)
+                if markers:
+                    return _Extracted(
+                        kind, clean, "text", {"markers": markers}, [], text)
             return _Extracted(kind, text, "text", None, [])
         if ptype == "reasoning":
             summary = _join_content_texts(payload.get("summary"))
             body = _join_content_texts(payload.get("content"))
+            reasoning = _reasoning_projection(
+                source="response_item", summary=summary, body=body)
+            if reasoning is None:
+                return None
             text = "\n".join(p for p in (summary, body) if p)
-            return _Extracted("reasoning", text, "search_thinking", None, [])
+            return _Extracted(
+                "reasoning", text, "search_thinking", {"reasoning": reasoning}, [])
         if ptype in _RESPONSE_TOOL_CALLS:
             name = payload.get("name") or ptype
             if ptype == "function_call":
@@ -231,20 +1217,55 @@ def _extract(record_type: str | None, payload: dict) -> _Extracted | None:
                 args = _stringify(payload.get("input") or payload.get("arguments"))
             text = f"{name}\n{args}" if args else str(name)
             detail = {"name": name, "args": args[:CODEX_TEXT_CAP]}
+            card = decode_tool_call_card(payload)
+            if card is None:
+                card = decode_secondary_tool_call_card(payload)
+            if card is not None:
+                detail["card"] = card
             return _Extracted("tool_call", text, "search_tool", detail, [])
         if ptype in _RESPONSE_TOOL_OUTPUTS:
-            body = _stringify(payload.get("output") or payload.get("tools"))
-            return _Extracted("tool_output", body, "search_tool", None, [])
+            value = payload["output"] if "output" in payload else payload.get("tools")
+            shaped = decode_tool_output_card(payload)
+            if shaped is not None:
+                card, _display_body = shaped
+                return _Extracted(
+                    "tool_output", _stringify(value), "search_tool", {"card": card}, [])
+            return _Extracted("tool_output", _stringify(value), "search_tool", None, [])
         return None  # unknown response_item subtype: version tolerance
     if record_type == "event_msg":
         if ptype in _EVENT_PROSE_KIND:
             kind = _EVENT_PROSE_KIND[ptype]
             text = _stringify(payload.get("message") or payload.get("text"))
+            role = "user" if kind == "user" else "assistant"
+            meta = _classify_injected_message(role, text)
+            if meta is not None:
+                return _Extracted("meta", text, "text", meta, [])
             column = "search_thinking" if kind == "reasoning" else "text"
+            if kind == "reasoning":
+                title_like = _REASONING_TITLE_RE.fullmatch(text.strip()) is not None
+                reasoning = _reasoning_projection(
+                    source="agent_reasoning",
+                    summary=text if title_like else "",
+                    body="" if title_like else text)
+                if reasoning is None:
+                    return None
+                return _Extracted(
+                    kind, text, column, {"reasoning": reasoning}, [])
+            if kind == "assistant":
+                clean, markers = _segment_harness_markers(text)
+                if markers:
+                    return _Extracted(
+                        kind, clean, column, {"markers": markers}, [], text)
             return _Extracted(kind, text, column, None, [])
         if ptype in _EVENT_CARD_TYPES:
             text, touches = _event_card(ptype, payload)
             detail = {"event": ptype}
+            if ptype in {"task_started", "task_complete"}:
+                detail.update(_lifecycle_projection(payload))
+            patch_card = decode_patch_event_card(payload)
+            card = patch_card or decode_secondary_event_card(payload)
+            if card is not None:
+                detail["card"] = card
             return _Extracted("event", text, "search_tool", detail, touches)
         return None  # token_count (accounting) + unknown event types: no row
     # session_meta / turn_context / unknown record types: no normalized row
@@ -268,8 +1289,12 @@ def _event_card(ptype: str, payload: dict) -> tuple[str, list[tuple[str, str]]]:
         text = " ".join(["patch_apply", *paths]) if paths else "patch_apply"
     elif ptype == "mcp_tool_call_end":
         invocation = payload.get("invocation")
-        name = invocation.get("name") if isinstance(invocation, dict) else None
-        text = f"mcp_tool_call {name}" if name else "mcp_tool_call"
+        server = invocation.get("server") if isinstance(invocation, dict) else None
+        tool = ((invocation.get("tool") or invocation.get("name"))
+                if isinstance(invocation, dict) else None)
+        identity = "/".join(
+            value for value in (server, tool) if isinstance(value, str) and value)
+        text = f"mcp_tool_call {identity}" if identity else "mcp_tool_call"
     elif ptype == "web_search_end":
         query = _stringify(payload.get("query"))
         text = f"web_search {query}".strip()
@@ -278,6 +1303,75 @@ def _event_card(ptype: str, payload: dict) -> tuple[str, list[tuple[str, str]]]:
     return text, touches
 
 
+def infer_codex_event_turns(
+    events: Iterable[Any], *, initial_turn: str | None = None
+) -> tuple[list[str | None], str | None]:
+    """Infer each physical event's logical turn from native lifecycle anchors.
+
+    ``turn_context``/``task_started`` establish a forward turn.  A resumed
+    segment can instead begin with response rows and expose its first native
+    proof only on a later patch/task-complete record; in that case only the
+    unanchored prefix since the latest ``session_meta`` is backfilled.  Distinct
+    explicit anchors are never merged.
+    """
+    materialized = list(events)
+    turns: list[str | None] = [None] * len(materialized)
+    current = initial_turn
+    segment_start = 0
+    for index, event in enumerate(materialized):
+        record_type = getattr(event, "record_type", None)
+        try:
+            obj = json.loads(getattr(event, "payload_json", "") or "{}")
+        except (json.JSONDecodeError, TypeError):
+            obj = {}
+        payload = obj.get("payload") if isinstance(obj.get("payload"), dict) else {}
+        ptype = payload.get("type")
+        if record_type == "session_meta":
+            current = None
+            segment_start = index + 1
+            continue
+        explicit = getattr(event, "turn_id", None)
+        if explicit is None and record_type == "turn_context":
+            candidate = payload.get("turn_id")
+            explicit = candidate if isinstance(candidate, str) and candidate else None
+        if explicit is not None:
+            if current is None and _is_late_turn_anchor(
+                    record_type, ptype, explicit):
+                for prior in range(segment_start, index):
+                    if turns[prior] is None:
+                        turns[prior] = explicit
+            current = explicit
+            turns[index] = explicit
+        else:
+            turns[index] = current
+    return turns, current
+
+
+def _is_late_turn_anchor(
+    record_type: str | None, payload_type: str | None, explicit_turn: str | None,
+) -> bool:
+    """Whether a native turn id can prove a preceding unanchored prefix."""
+    return (
+        explicit_turn is not None
+        and record_type != "turn_context"
+        and payload_type != "task_started"
+    )
+
+
+def codex_event_is_late_turn_anchor(event: Any) -> bool:
+    """Public predicate shared by delta persistence and turn inference."""
+    try:
+        obj = json.loads(getattr(event, "payload_json", "") or "{}")
+    except (json.JSONDecodeError, TypeError):
+        obj = {}
+    payload = obj.get("payload") if isinstance(obj.get("payload"), dict) else {}
+    return _is_late_turn_anchor(
+        getattr(event, "record_type", None),
+        payload.get("type"),
+        getattr(event, "turn_id", None),
+    )
+
+
 def normalize_codex_events(
     events: Iterable[Any], *, initial: CodexStickyState
 ) -> CodexNormalizationResult:
@@ -285,10 +1379,13 @@ def normalize_codex_events(
     rows + file touches, replaying sticky turn/model state seeded from ``initial``.
 
     Returns the terminal sticky state for persistence to codex_session_files."""
+    materialized = list(events)
+    inferred_turns, terminal_turn = infer_codex_event_turns(
+        materialized, initial_turn=initial.turn_id)
     sticky = CodexStickyState(turn_id=initial.turn_id, model=initial.model)
     rows: list[CodexNormalizedRow] = []
     touches: list[CodexFileTouch] = []
-    for event in events:
+    for event_index, event in enumerate(materialized):
         record_type = getattr(event, "record_type", None)
         if record_type == "session_meta":
             sticky.turn_id = None
@@ -301,7 +1398,9 @@ def normalize_codex_events(
         payload = obj.get("payload") if isinstance(obj.get("payload"), dict) else {}
         if record_type == "turn_context":
             own = getattr(event, "turn_id", None)
-            sticky.turn_id = own if own is not None else payload.get("turn_id")
+            candidate_turn = own if own is not None else payload.get("turn_id")
+            if isinstance(candidate_turn, str) and candidate_turn:
+                sticky.turn_id = candidate_turn
             model = payload.get("model")
             sticky.model = model if isinstance(model, str) and model else None
             continue
@@ -313,9 +1412,10 @@ def normalize_codex_events(
         extracted = _extract(record_type, payload)
         if extracted is None:
             continue
-        own_turn = getattr(event, "turn_id", None)
-        eff_turn = own_turn if own_turn is not None else sticky.turn_id
+        eff_turn = inferred_turns[event_index]
         text_full = extracted.content_text or ""
+        identity_full = (extracted.identity_text
+                         if extracted.identity_text is not None else text_full)
         capped, truncated = _cap(text_full)
         text_col = ""
         search_tool = ""
@@ -332,6 +1432,11 @@ def normalize_codex_events(
             detail["truncated"] = True
         source_path = getattr(event, "source_path", "")
         line_offset = getattr(event, "line_offset", 0)
+        effective_call_id = getattr(event, "call_id", None)
+        if not effective_call_id and record_type == "response_item":
+            candidate = payload.get("id")
+            if isinstance(candidate, str) and candidate:
+                effective_call_id = candidate
         rows.append(CodexNormalizedRow(
             conversation_key=conversation_key,
             source_root_key=getattr(event, "source_root_key", None) or "",
@@ -339,14 +1444,14 @@ def normalize_codex_events(
             line_offset=line_offset,
             timestamp_utc=getattr(event, "timestamp_utc", None),
             turn_id=eff_turn,
-            call_id=getattr(event, "call_id", None),
+            call_id=effective_call_id,
             kind=extracted.kind,
             event_type=getattr(event, "event_type", None),
             record_family="event_msg" if record_type == "event_msg" else "response_item",
             model=sticky.model,
             text=text_col,
-            content_digest=content_digest(text_full),
-            content_len=content_len(text_full),
+            content_digest=content_digest(identity_full),
+            content_len=content_len(identity_full),
             detail_json=_canonical_json(detail) if detail else None,
             search_tool=search_tool,
             search_thinking=search_thinking,
@@ -361,7 +1466,7 @@ def normalize_codex_events(
             ))
     return CodexNormalizationResult(
         rows=rows, touches=touches,
-        terminal=CodexStickyState(turn_id=sticky.turn_id, model=sticky.model),
+        terminal=CodexStickyState(turn_id=terminal_turn, model=sticky.model),
     )
 
 
@@ -383,7 +1488,7 @@ def _pair_mirrors_impl(
     # member (three identical event copies never collapse into one).
     turned: dict[tuple, dict[str, list[int]]] = {}
     for i, row in enumerate(rows):
-        if row.kind not in _PROSE_KINDS or row.turn_id is None:
+        if row.kind not in _MIRROR_KINDS or row.turn_id is None:
             continue
         key = (row.turn_id, row.kind, row.content_digest, row.content_len)
         group = turned.setdefault(key, {"R": [], "E": []})
@@ -399,7 +1504,7 @@ def _pair_mirrors_impl(
     last_prose_by_kind: dict[str, int] = {}
     paired_response: set[int] = set()
     for j, row in enumerate(rows):
-        if row.kind not in _PROSE_KINDS or row.turn_id is not None:
+        if row.kind not in _MIRROR_KINDS or row.turn_id is not None:
             continue
         prev_idx = last_prose_by_kind.get(row.kind)
         if row.record_family == "event_msg" and prev_idx is not None:
@@ -446,7 +1551,219 @@ def _item_pos(row: CodexNormalizedRow) -> tuple:
     return (row.timestamp_utc or "", row.source_path, row.line_offset)
 
 
-def canonical_items(rows: list[CodexNormalizedRow]) -> list[dict]:
+def _row_card(row: CodexNormalizedRow) -> dict | None:
+    if not row.detail_json:
+        return None
+    try:
+        detail = json.loads(row.detail_json)
+    except (json.JSONDecodeError, TypeError):
+        return None
+    card = detail.get("card") if isinstance(detail, dict) else None
+    return card if isinstance(card, dict) else None
+
+
+def _fold_patch_completion_items(items: list[dict]) -> list[dict]:
+    """Fold only proven ``patch_apply_end`` items into their owning response.
+
+    Same-id ownership wins when unique. Current nested ``tools.apply_patch``
+    events use an inner id, so the second proof is a unique physical bracket in
+    one source file: call < event < that call's uniquely-owned output. Ambiguous
+    or repeated shapes stay separate and inspectable.
+    """
+    patch_calls: list[
+        tuple[int, CodexNormalizedRow, CodexNormalizedRow | None, int]
+    ] = []
+    for item_index, item in enumerate(items):
+        if item["klass"] != "response":
+            continue
+        owner_count: dict[str, int] = {}
+        for row in item["rows"]:
+            if row.kind == "tool_call" and row.call_id:
+                owner_count[row.call_id] = owner_count.get(row.call_id, 0) + 1
+        for row in item["rows"]:
+            card = _row_card(row)
+            if not (row.kind == "tool_call" and isinstance(card, dict)
+                    and card.get("type") == "patch"):
+                continue
+            outputs = [
+                candidate for candidate in item["rows"]
+                if candidate.kind == "tool_output" and row.call_id
+                and candidate.call_id == row.call_id
+                and owner_count.get(row.call_id) == 1
+                and _item_pos(row) < _item_pos(candidate)
+            ]
+            patch_calls.append((
+                item_index, row, outputs[0] if len(outputs) == 1 else None,
+                owner_count.get(row.call_id, 0),
+            ))
+
+    matched_calls: set[tuple[str, int]] = set()
+    remove_items: set[int] = set()
+    for event_index, event_item in enumerate(items):
+        if event_item["klass"] != "event" or len(event_item["rows"]) != 1:
+            continue
+        event_row = event_item["rows"][0]
+        event_card = _row_card(event_row)
+        if not (isinstance(event_card, dict)
+                and event_card.get("source") == "patch_apply_end"):
+            continue
+        same_id = [
+            candidate for candidate in patch_calls
+            if event_row.call_id and candidate[1].call_id == event_row.call_id
+            and candidate[3] == 1
+            and candidate[1].turn_id == event_row.turn_id
+            and (candidate[1].source_path, candidate[1].line_offset) not in matched_calls
+        ]
+        candidates = same_id if len(same_id) == 1 else []
+        if not candidates:
+            candidates = [
+                candidate for candidate in patch_calls
+                if candidate[2] is not None
+                and candidate[1].turn_id == event_row.turn_id
+                and candidate[1].source_path == event_row.source_path
+                and candidate[2].source_path == event_row.source_path
+                and candidate[1].line_offset < event_row.line_offset < candidate[2].line_offset
+                and (candidate[1].source_path, candidate[1].line_offset) not in matched_calls
+            ]
+        if len(candidates) != 1:
+            continue
+        owner_index, call_row, _output_row, _owner_count = candidates[0]
+        owner = items[owner_index]
+        owner["rows"].append(event_row)
+        owner["rows"].sort(key=_item_pos)
+        owner.setdefault("folded_items", []).append(event_item)
+        matched_calls.add((call_row.source_path, call_row.line_offset))
+        remove_items.add(event_index)
+    return [item for index, item in enumerate(items) if index not in remove_items]
+
+
+def _fold_secondary_completion_items(items: list[dict]) -> list[dict]:
+    """Fold web/MCP end events only onto one exact same-turn call id.
+
+    No name, adjacency, or basename inference is allowed: a reused, empty, or
+    unmatched id leaves the native completion visible as its own event card.
+    """
+    calls_by_id: dict[str, list[tuple[int, CodexNormalizedRow]]] = {}
+    for item_index, item in enumerate(items):
+        if item["klass"] != "response":
+            continue
+        for row in item["rows"]:
+            if row.kind == "tool_call" and row.call_id:
+                calls_by_id.setdefault(row.call_id, []).append((item_index, row))
+    remove_items: set[int] = set()
+    matched_calls: set[tuple[str, int]] = set()
+    for event_index, event_item in enumerate(items):
+        if event_item["klass"] != "event" or len(event_item["rows"]) != 1:
+            continue
+        event_row = event_item["rows"][0]
+        event_card = _row_card(event_row)
+        if not (event_row.call_id and isinstance(event_card, dict)
+                and event_card.get("type") in {
+                    "web_search_completion", "mcp_completion"}):
+            continue
+        candidates = [
+            (owner_index, call_row)
+            for owner_index, call_row in calls_by_id.get(event_row.call_id, [])
+            if call_row.turn_id == event_row.turn_id
+            and (call_row.source_path, call_row.line_offset) not in matched_calls
+        ]
+        if event_card.get("type") == "web_search_completion":
+            candidates = [
+                candidate for candidate in candidates
+                if (_parse_row_detail(candidate[1]) or {}).get("name")
+                == "web_search_call"
+            ]
+        if len(candidates) != 1:
+            continue
+        owner_index, call_row = candidates[0]
+        owner = items[owner_index]
+        owner["rows"].append(event_row)
+        owner["rows"].sort(key=_item_pos)
+        owner.setdefault("folded_items", []).append(event_item)
+        matched_calls.add((call_row.source_path, call_row.line_offset))
+        remove_items.add(event_index)
+    return [item for index, item in enumerate(items) if index not in remove_items]
+
+
+def _parse_row_detail(row: CodexNormalizedRow) -> dict | None:
+    if not row.detail_json:
+        return None
+    try:
+        detail = json.loads(row.detail_json)
+    except (json.JSONDecodeError, TypeError):
+        return None
+    return detail if isinstance(detail, dict) else None
+
+
+def _fold_lifecycle_items(items: list[dict]) -> list[dict]:
+    """Fold only closed, uniquely-owned redundant task lifecycle events.
+
+    The physical rows remain retained and become member-key aliases. A completion
+    is redundant only when its last message is empty or digest/length-identical to
+    a real assistant row in the one owning logical response. Unknown fields,
+    errors, unmatched/ambiguous events, and unique messages stay standalone.
+    """
+    responses: dict[str, list[tuple[int, dict]]] = {}
+    events: dict[tuple[str, str], list[tuple[int, dict, CodexNormalizedRow, dict]]] = {}
+    for index, item in enumerate(items):
+        turn_id = item.get("turn_id")
+        if item["klass"] == "response" and turn_id:
+            responses.setdefault(turn_id, []).append((index, item))
+        if item["klass"] != "event" or len(item["rows"]) != 1 or not turn_id:
+            continue
+        row = item["rows"][0]
+        detail = _parse_row_detail(row)
+        lifecycle = detail.get("lifecycle") if isinstance(detail, dict) else None
+        if (isinstance(lifecycle, dict)
+                and lifecycle.get("event") in {"task_started", "task_complete"}):
+            events.setdefault((turn_id, lifecycle["event"]), []).append(
+                (index, item, row, detail))
+
+    remove: set[int] = set()
+    for turn_id, owners in responses.items():
+        if len(owners) != 1:
+            continue
+        _owner_index, owner = owners[0]
+        for event_name in ("task_started", "task_complete"):
+            candidates = events.get((turn_id, event_name), [])
+            if len(candidates) != 1:
+                continue
+            event_index, event_item, row, detail = candidates[0]
+            if detail.get("_lifecycle_foldable") is not True:
+                continue
+            if event_name == "task_complete":
+                message_len = detail.get("_lifecycle_message_len")
+                message_digest = detail.get("_lifecycle_message_digest")
+                if message_len:
+                    if not any(
+                        candidate.kind == "assistant"
+                        and candidate.content_len == message_len
+                        and candidate.content_digest == message_digest
+                        for candidate in owner["rows"]
+                    ):
+                        continue
+            lifecycle = detail["lifecycle"]
+            section = {
+                key: value for key, value in lifecycle.items()
+                if key not in {"schema_version", "event", "state", "message", "error"}
+            }
+            folded = owner.setdefault(
+                "lifecycle", {"schema_version": 1, "state": "started"})
+            folded["started" if event_name == "task_started" else "completed"] = section
+            if event_name == "task_complete":
+                folded["state"] = "completed"
+            owner["rows"].append(row)
+            owner["rows"].sort(key=_item_pos)
+            owner.setdefault("lifecycle_rows", []).append(row)
+            owner["lifecycle_rows"].sort(key=_item_pos)
+            owner.setdefault("folded_items", []).append(event_item)
+            remove.add(event_index)
+    return [item for index, item in enumerate(items) if index not in remove]
+
+
+def canonical_items(
+    rows: list[CodexNormalizedRow], *, fold_patch_completions: bool = True,
+) -> list[dict]:
     """Group KEPT (post-pairing) rows into canonical rendered items (§5.2).
 
     Response items bundle a turn's assistant/reasoning/tool rows; prompt/event/
@@ -460,6 +1777,9 @@ def canonical_items(rows: list[CodexNormalizedRow]) -> list[dict]:
         if row.turn_id is None:
             items.append({"klass": "unturned", "rows": [row], "turn_id": None,
                           "anchor_row": row})
+        elif row.kind == "meta":
+            items.append({"klass": "meta", "rows": [row], "turn_id": row.turn_id,
+                          "anchor_row": row})
         elif row.kind == "user":
             items.append({"klass": "prompt", "rows": [row], "turn_id": row.turn_id,
                           "anchor_row": row})
@@ -474,6 +1794,10 @@ def canonical_items(rows: list[CodexNormalizedRow]) -> list[dict]:
                 response_by_turn[row.turn_id] = item
                 items.append(item)
             item["rows"].append(row)
+    if fold_patch_completions:
+        items = _fold_patch_completion_items(items)
+        items = _fold_secondary_completion_items(items)
+    items = _fold_lifecycle_items(items)
     items.sort(key=lambda it: _item_pos(it["anchor_row"]))
     return items