hermes-agent/agent/message_sanitization.py

852 lines
36 KiB
Python

"""Message and tool-payload sanitization helpers.
Pure functions extracted from ``run_agent.py`` so the AIAgent module can
stay focused on the conversation loop. These walk OpenAI-format message
lists and structured payloads, repairing or stripping problematic
characters that would otherwise crash ``json.dumps`` inside the OpenAI
SDK or be rejected by upstream APIs.
All helpers are stateless and side-effect-free except for in-place
mutation of their input (where documented). Backward-compatible
re-exports from ``run_agent`` remain in place so existing imports
``from run_agent import _sanitize_surrogates`` keep working.
"""
from __future__ import annotations
import hashlib
import json
import logging
import re
from typing import Any
logger = logging.getLogger(__name__)
# Lone surrogate code points are invalid in UTF-8 and crash json.dumps
# inside the OpenAI SDK. Used by every surrogate-sanitization helper
# below as well as by run_agent and the CLI for paste-from-clipboard
# scrubbing.
_SURROGATE_RE = re.compile(r'[\ud800-\udfff]')
def _sanitize_surrogates(text: str) -> str:
"""Replace lone surrogate code points with U+FFFD (replacement character).
Surrogates are invalid in UTF-8 and will crash ``json.dumps()`` inside the
OpenAI SDK. This is a fast no-op when the text contains no surrogates.
"""
if _SURROGATE_RE.search(text):
return _SURROGATE_RE.sub('\ufffd', text)
return text
def _sanitize_structure_surrogates(payload: Any) -> bool:
"""Replace surrogate code points in nested dict/list payloads in-place.
Mirror of ``_sanitize_structure_non_ascii`` but for surrogate recovery.
Used to scrub nested structured fields (e.g. ``reasoning_details`` — an
array of dicts with ``summary``/``text`` strings) that flat per-field
checks don't reach. Returns True if any surrogates were replaced.
"""
found = False
def _walk(node):
nonlocal found
if isinstance(node, dict):
for key, value in node.items():
if isinstance(value, str):
if _SURROGATE_RE.search(value):
node[key] = _SURROGATE_RE.sub('\ufffd', value)
found = True
elif isinstance(value, (dict, list)):
_walk(value)
elif isinstance(node, list):
for idx, value in enumerate(node):
if isinstance(value, str):
if _SURROGATE_RE.search(value):
node[idx] = _SURROGATE_RE.sub('\ufffd', value)
found = True
elif isinstance(value, (dict, list)):
_walk(value)
_walk(payload)
return found
def _sanitize_messages_surrogates(messages: list) -> bool:
"""Sanitize surrogate characters from all string content in a messages list.
Walks message dicts in-place. Returns True if any surrogates were found
and replaced, False otherwise. Covers content/text, name, tool call
metadata/arguments, AND any additional string or nested structured fields
(``reasoning``, ``reasoning_content``, ``reasoning_details``, etc.) so
retries don't fail on a non-content field. Byte-level reasoning models
(xiaomi/mimo, kimi, glm) can emit lone surrogates in reasoning output
that flow through to ``api_messages["reasoning_content"]`` on the next
turn and crash json.dumps inside the OpenAI SDK.
"""
found = False
for msg in messages:
if not isinstance(msg, dict):
continue
content = msg.get("content")
if isinstance(content, str) and _SURROGATE_RE.search(content):
msg["content"] = _SURROGATE_RE.sub('\ufffd', content)
found = True
elif isinstance(content, list):
for part in content:
if isinstance(part, dict):
text = part.get("text")
if isinstance(text, str) and _SURROGATE_RE.search(text):
part["text"] = _SURROGATE_RE.sub('\ufffd', text)
found = True
name = msg.get("name")
if isinstance(name, str) and _SURROGATE_RE.search(name):
msg["name"] = _SURROGATE_RE.sub('\ufffd', name)
found = True
tool_calls = msg.get("tool_calls")
if isinstance(tool_calls, list):
for tc in tool_calls:
if not isinstance(tc, dict):
continue
tc_id = tc.get("id")
if isinstance(tc_id, str) and _SURROGATE_RE.search(tc_id):
tc["id"] = _SURROGATE_RE.sub('\ufffd', tc_id)
found = True
fn = tc.get("function")
if isinstance(fn, dict):
fn_name = fn.get("name")
if isinstance(fn_name, str) and _SURROGATE_RE.search(fn_name):
fn["name"] = _SURROGATE_RE.sub('\ufffd', fn_name)
found = True
fn_args = fn.get("arguments")
if isinstance(fn_args, str) and _SURROGATE_RE.search(fn_args):
fn["arguments"] = _SURROGATE_RE.sub('\ufffd', fn_args)
found = True
# Walk any additional string / nested fields (reasoning,
# reasoning_content, reasoning_details, etc.) — surrogates from
# byte-level reasoning models (xiaomi/mimo, kimi, glm) can lurk
# in these fields and aren't covered by the per-field checks above.
# Matches _sanitize_messages_non_ascii's coverage (PR #10537).
for key, value in msg.items():
if key in {"content", "name", "tool_calls", "role"}:
continue
if isinstance(value, str):
if _SURROGATE_RE.search(value):
msg[key] = _SURROGATE_RE.sub('\ufffd', value)
found = True
elif isinstance(value, (dict, list)):
if _sanitize_structure_surrogates(value):
found = True
return found
def _escape_invalid_chars_in_json_strings(raw: str) -> str:
"""Escape unescaped control chars inside JSON string values.
Walks the raw JSON character-by-character, tracking whether we are
inside a double-quoted string. Inside strings, replaces literal
control characters (0x00-0x1F) that aren't already part of an escape
sequence with their ``\\uXXXX`` equivalents. Pass-through for everything
else.
Ported from #12093 — complements the other repair passes in
``_repair_tool_call_arguments`` when ``json.loads(strict=False)`` is
not enough (e.g. llama.cpp backends that emit literal apostrophes or
tabs alongside other malformations).
"""
out: list[str] = []
in_string = False
i = 0
n = len(raw)
while i < n:
ch = raw[i]
if in_string:
if ch == "\\" and i + 1 < n:
# Already-escaped char — pass through as-is
out.append(ch)
out.append(raw[i + 1])
i += 2
continue
if ch == '"':
in_string = False
out.append(ch)
elif ord(ch) < 0x20:
out.append(f"\\u{ord(ch):04x}")
else:
out.append(ch)
else:
if ch == '"':
in_string = True
out.append(ch)
i += 1
return "".join(out)
def _repair_tool_call_arguments(raw_args: str, tool_name: str = "?") -> str:
"""Attempt to repair malformed tool_call argument JSON.
Models like GLM-5.1 via Ollama can produce truncated JSON, trailing
commas, Python ``None``, etc. The API proxy rejects these with HTTP 400
"invalid tool call arguments". This function applies common repairs;
if all fail it returns ``"{}"`` so the request succeeds (better than
crashing the session). All repairs are logged at WARNING level.
"""
raw_stripped = raw_args.strip() if isinstance(raw_args, str) else ""
# Fast-path: empty / whitespace-only -> empty object
if not raw_stripped:
logger.warning("Sanitized empty tool_call arguments for %s", tool_name)
return "{}"
# Python-literal None -> normalise to {}
if raw_stripped == "None":
logger.warning("Sanitized Python-None tool_call arguments for %s", tool_name)
return "{}"
# Repair pass 0: llama.cpp backends sometimes emit literal control
# characters (tabs, newlines) inside JSON string values. json.loads
# with strict=False accepts these and lets us re-serialise the
# result into wire-valid JSON without any string surgery. This is
# the most common local-model repair case (#12068).
try:
parsed = json.loads(raw_stripped, strict=False)
reserialised = json.dumps(parsed, separators=(",", ":"))
if reserialised != raw_stripped:
logger.warning(
"Repaired unescaped control chars in tool_call arguments for %s",
tool_name,
)
return reserialised
except (json.JSONDecodeError, TypeError, ValueError):
pass
# Attempt common JSON repairs
fixed = raw_stripped
# 1. Strip trailing commas before } or ]
fixed = re.sub(r',\s*([}\]])', r'\1', fixed)
# 2. Close unclosed structures
open_curly = fixed.count('{') - fixed.count('}')
open_bracket = fixed.count('[') - fixed.count(']')
if open_curly > 0:
fixed += '}' * open_curly
if open_bracket > 0:
fixed += ']' * open_bracket
# 3. Remove excess closing braces/brackets (bounded to 50 iterations)
for _ in range(50):
try:
json.loads(fixed)
break
except json.JSONDecodeError:
if fixed.endswith('}') and fixed.count('}') > fixed.count('{'):
fixed = fixed[:-1]
elif fixed.endswith(']') and fixed.count(']') > fixed.count('['):
fixed = fixed[:-1]
else:
break
try:
json.loads(fixed)
logger.warning(
"Repaired malformed tool_call arguments for %s: %s%s",
tool_name, raw_stripped[:80], fixed[:80],
)
return fixed
except json.JSONDecodeError:
pass
# Repair pass 4: escape unescaped control chars inside JSON strings,
# then retry. Catches cases where strict=False alone fails because
# other malformations are present too.
try:
escaped = _escape_invalid_chars_in_json_strings(fixed)
if escaped != fixed:
json.loads(escaped)
logger.warning(
"Repaired control-char-laced tool_call arguments for %s: %s%s",
tool_name, raw_stripped[:80], escaped[:80],
)
return escaped
except (json.JSONDecodeError, TypeError, ValueError):
pass
# Last resort: replace with empty object so the API request doesn't
# crash the entire session.
logger.warning(
"Unrepairable tool_call arguments for %s"
"replaced with empty object (was: %s)",
tool_name, raw_stripped[:80],
)
return "{}"
def close_interrupted_tool_sequence(messages: list, final_response: Any = None) -> bool:
"""Append a synthetic assistant turn when an interrupted tail is a tool result.
A turn cut short by ``/stop`` can leave the transcript ending on a raw
``tool`` message (a tool finished, or its execution was cancelled, but the
model never streamed a closing assistant turn). Persisting that tail means
the next user message lands as ``… tool → user`` — a role-alternation
violation that strict providers (Gemini, Claude) react to by hallucinating
a continuation of the user's message and ignoring prior context, which
reads to the user as "lost context" (#48879).
``finalize_turn`` closes this on the happy interrupt path, but the
retry/backoff/error interrupt aborts in ``conversation_loop`` ``return``
early and never reach it — this shared helper closes the sequence on all of
them. ``final_response`` is usually empty on an interrupt, so an explicit
placeholder is used rather than an empty-content assistant turn.
Mutates ``messages`` in place. Returns True if a closing turn was appended.
"""
if not messages:
return False
last = messages[-1]
if not isinstance(last, dict) or last.get("role") != "tool":
return False
text = final_response if isinstance(final_response, str) else ""
messages.append({
"role": "assistant",
"content": text.strip() or "Operation interrupted.",
})
return True
def _strip_non_ascii(text: str) -> str:
"""Remove non-ASCII characters, replacing with closest ASCII equivalent or removing.
Used as a last resort when the system encoding is ASCII and can't handle
any non-ASCII characters (e.g. LANG=C on Chromebooks).
"""
return text.encode('ascii', errors='ignore').decode('ascii')
def _sanitize_messages_non_ascii(messages: list) -> bool:
"""Strip non-ASCII characters from all string content in a messages list.
This is a last-resort recovery for systems with ASCII-only encoding
(LANG=C, Chromebooks, minimal containers). Returns True if any
non-ASCII content was found and sanitized.
"""
found = False
for msg in messages:
if not isinstance(msg, dict):
continue
# Sanitize content (string)
content = msg.get("content")
if isinstance(content, str):
sanitized = _strip_non_ascii(content)
if sanitized != content:
msg["content"] = sanitized
found = True
elif isinstance(content, list):
for part in content:
if isinstance(part, dict):
text = part.get("text")
if isinstance(text, str):
sanitized = _strip_non_ascii(text)
if sanitized != text:
part["text"] = sanitized
found = True
# Sanitize name field (can contain non-ASCII in tool results)
name = msg.get("name")
if isinstance(name, str):
sanitized = _strip_non_ascii(name)
if sanitized != name:
msg["name"] = sanitized
found = True
# Sanitize tool_calls
tool_calls = msg.get("tool_calls")
if isinstance(tool_calls, list):
for tc in tool_calls:
if isinstance(tc, dict):
fn = tc.get("function", {})
if isinstance(fn, dict):
fn_args = fn.get("arguments")
if isinstance(fn_args, str):
sanitized = _strip_non_ascii(fn_args)
if sanitized != fn_args:
fn["arguments"] = sanitized
found = True
# Sanitize any additional top-level string fields (e.g. reasoning_content)
for key, value in msg.items():
if key in {"content", "name", "tool_calls", "role"}:
continue
if isinstance(value, str):
sanitized = _strip_non_ascii(value)
if sanitized != value:
msg[key] = sanitized
found = True
return found
def _sanitize_tools_non_ascii(tools: list) -> bool:
"""Strip non-ASCII characters from tool payloads in-place."""
return _sanitize_structure_non_ascii(tools)
def _strip_images_from_messages(messages: list) -> bool:
"""Remove image_url content parts from all messages in-place.
Called when a server signals it does not support images (e.g.
"Only 'text' content type is supported."). Mutates messages so the
next API call sends text only.
Preserves message alternation invariants:
* ``tool``-role messages whose content was entirely images are replaced
with a plaintext placeholder, NOT deleted — deleting them would leave
the paired ``tool_call_id`` on the prior assistant message unmatched,
which providers reject with HTTP 400.
* Non-tool messages whose content becomes empty are dropped. In
practice this only hits synthetic image-only user messages appended
for attachment delivery; real user turns always include text.
Returns True if any image parts were removed.
"""
found = False
to_delete = []
for i, msg in enumerate(messages):
if not isinstance(msg, dict):
continue
content = msg.get("content")
if not isinstance(content, list):
continue
new_parts = []
for part in content:
if isinstance(part, dict) and part.get("type") in {"image_url", "image", "input_image"}:
found = True
else:
new_parts.append(part)
if len(new_parts) < len(content):
if new_parts:
msg["content"] = new_parts
elif msg.get("role") == "tool":
# Preserve tool_call_id linkage — providers require every
# assistant tool_call to have a matching tool response.
msg["content"] = "[image content removed — server does not support images]"
else:
# Synthetic image-only user/assistant message with no text;
# safe to drop.
to_delete.append(i)
for i in reversed(to_delete):
del messages[i]
return found
def _sanitize_structure_non_ascii(payload: Any) -> bool:
"""Strip non-ASCII characters from nested dict/list payloads in-place."""
found = False
def _walk(node):
nonlocal found
if isinstance(node, dict):
for key, value in node.items():
if isinstance(value, str):
sanitized = _strip_non_ascii(value)
if sanitized != value:
node[key] = sanitized
found = True
elif isinstance(value, (dict, list)):
_walk(value)
elif isinstance(node, list):
for idx, value in enumerate(node):
if isinstance(value, str):
sanitized = _strip_non_ascii(value)
if sanitized != value:
node[idx] = sanitized
found = True
elif isinstance(value, (dict, list)):
_walk(value)
_walk(payload)
return found
__all__ = [
"_SURROGATE_RE",
"close_interrupted_tool_sequence",
"_sanitize_surrogates",
"_sanitize_structure_surrogates",
"_sanitize_messages_surrogates",
"_escape_invalid_chars_in_json_strings",
"_repair_tool_call_arguments",
"_strip_non_ascii",
"_sanitize_messages_non_ascii",
"_sanitize_tools_non_ascii",
"_strip_images_from_messages",
"_sanitize_structure_non_ascii",
# call_id policy owners (F4 consolidation)
"deterministic_call_id",
"coalesce_tool_call_id",
"uniquify_tool_call_ids",
# reasoning_content policy owners (F4 consolidation)
"reasoning_echo_family",
"matches_reasoning_echo_family",
"needs_reasoning_echo",
"apply_reasoning_content_policy",
"reapply_reasoning_echo",
]
# ---------------------------------------------------------------------------
# call_id policy — single owner (audit F4, incident chain I4)
# ---------------------------------------------------------------------------
#
# Three forked policy sites converged here:
# * agent/codex_responses_adapter.py `_deterministic_call_id` — hash
# synthesis when a provider omits call_id (fa3ab2ffd0 → e45f2b39e2).
# * run_agent.AIAgent._get_tool_call_id_static — `call_id or id`
# coalescing for dicts and SDK objects.
# * run_agent.AIAgent._uniquify_tool_call_ids — duplicate-id repair with
# deterministic `_d<n>` suffixes (#58327 loss class).
#
# NOT consolidated (different scheme on purpose):
# agent/transports/codex_event_projector._deterministic_call_id maps codex
# app-server ITEM ids (`codex_<type>_<item_id>`), not chat tool-call
# content; merging the two would change ids and invalidate prompt caches.
#
# HARD INVARIANT: everything here must stay deterministic (never uuid4) and
# byte-identical for existing inputs — these ids feed prompt-cache prefixes.
def deterministic_call_id(fn_name: str, arguments: str, index: int = 0) -> str:
"""Generate a deterministic call_id from tool call content.
Used as a fallback when the API doesn't provide a call_id.
Deterministic IDs prevent cache invalidation — random UUIDs would
make every API call's prefix unique, breaking OpenAI's prompt cache.
"""
seed = f"{fn_name}:{arguments}:{index}"
digest = hashlib.sha256(seed.encode("utf-8", errors="replace")).hexdigest()[:12]
return f"call_{digest}"
def coalesce_tool_call_id(tc: Any) -> str:
"""Extract the effective call ID from a tool_call entry (dict or object).
Single owner for the ``call_id or id`` coalescing rule: Codex Responses
tool calls carry ``call_id`` (authoritative pairing key), Chat
Completions ones carry ``id`` only. Returns ``""`` when neither is set.
"""
if isinstance(tc, dict):
return (tc.get("call_id", "") or tc.get("id", "") or "").strip()
return (getattr(tc, "call_id", "") or getattr(tc, "id", "") or "").strip()
def uniquify_tool_call_ids(tool_calls: list) -> list:
"""Ensure every tool call in a single assistant turn has a distinct id.
Some models/providers reuse one call id across different calls in a
single batch (observed with native Kimi Responses replays, Ollama-
compatible endpoints, and degraded models at long context; same bug
class as openclaw/openclaw#110518 / #110956). Duplicate ids are lossy
downstream: the pre-API sanitizer keeps only the first call/result
pair per id (#58327), so the later call's result silently vanishes
from every replayed payload, and strict providers (Anthropic
tool_use, DeepSeek) reject duplicate ids outright.
The first occurrence keeps its id; later collisions get a
deterministic ``<id>_d<n>`` suffix — never a random UUID, which would
break prompt-cache prefix stability across replays. Mutates the
entries in place (SDK models / SimpleNamespace / dicts) and returns
the same list. Blank/missing ids are left for the deterministic
fallback in ``build_assistant_message``.
"""
seen: set = set()
for tc in tool_calls or []:
# Same coalescing rule as ``coalesce_tool_call_id`` but tolerant of
# non-string ids (degraded models can emit ints/None here).
if isinstance(tc, dict):
raw = tc.get("call_id") or tc.get("id") or ""
else:
raw = getattr(tc, "call_id", None) or getattr(tc, "id", None) or ""
raw = raw.strip() if isinstance(raw, str) else ""
if not raw:
continue
# Composite Responses ids ("call_x|fc_y") collide on the call
# half — that's the pairing key providers enforce per turn.
cid = raw.split("|", 1)[0]
if not cid:
continue
if cid not in seen:
seen.add(cid)
continue
n = 2
new_id = f"{cid}_d{n}"
while new_id in seen:
n += 1
new_id = f"{cid}_d{n}"
seen.add(new_id)
def _renamed(value):
# Preserve a composite id's response-item half so the
# provider's real fc_/item id survives the rename.
if isinstance(value, str) and "|" in value:
return f"{new_id}|{value.split('|', 1)[1]}"
return new_id
try:
if isinstance(tc, dict):
if tc.get("id"):
tc["id"] = _renamed(tc["id"])
else:
tc["id"] = new_id
if tc.get("call_id"):
tc["call_id"] = new_id
else:
tc.id = _renamed(getattr(tc, "id", None))
if getattr(tc, "call_id", None):
tc.call_id = new_id
except Exception:
logger.warning(
"Could not uniquify duplicate tool call id %s", cid
)
continue
_fn = tc.get("function") if isinstance(tc, dict) else getattr(tc, "function", None)
_fn_name = (_fn.get("name") if isinstance(_fn, dict) else getattr(_fn, "name", None)) or "?"
logger.warning(
"Model reused tool call id %s within one turn; renamed the "
"duplicate to %s (tool=%s) to keep call/result pairing "
"lossless.", cid, new_id, _fn_name,
)
return tool_calls
# ---------------------------------------------------------------------------
# reasoning_content policy — single owner (audit F4)
# ---------------------------------------------------------------------------
#
# The strip-vs-repad decision was previously forked across the wire files in
# separate incident commits (2b3a4f0af8 strip for strict providers,
# b5495db701 re-pad for require-side, 94b3131be7/9a9f8a6d99 kimi pad). The
# POLICY — which provider direction gets which treatment — lives here as one
# rule table + apply functions; adapters keep only SYNTAX mapping (e.g.
# anthropic_adapter turning reasoning_content into a thinking block).
#
# Direction table:
# require-side (echo-back enforced; replays 400 without the field):
# kimi — provider kimi-coding/kimi-coding-cn, or host api.kimi.com /
# moonshot.ai / moonshot.cn. Host-driven on purpose:
# aggregators re-exporting kimi models reject the echo.
# deepseek — provider "deepseek", model contains "deepseek", or host
# api.deepseek.com (#15250; V4 rejects empty-string pads,
# hence the " " single-space pad, #17341).
# mimo — provider "xiaomi", model contains "mimo", or host
# *.xiaomimimo.com.
# strict side (field rejected with 400/422 "Extra inputs are not
# permitted"): everyone else — Mistral, Cerebras, Groq, SambaNova, …
# (#45655). Strip the key entirely, even a single-space pad.
_REASONING_ECHO_RULES: tuple = (
# (family, exact providers (raw), exact providers (lowered),
# model substrings (lowered), base_url hosts)
("kimi", frozenset({"kimi-coding", "kimi-coding-cn"}), frozenset(), (),
("api.kimi.com", "moonshot.ai", "moonshot.cn")),
("deepseek", frozenset(), frozenset({"deepseek"}), ("deepseek",),
("api.deepseek.com",)),
("mimo", frozenset(), frozenset({"xiaomi"}), ("mimo",),
("api.xiaomimimo.com", "xiaomimimo.com")),
)
def _family_rule(family: str) -> tuple:
for rule in _REASONING_ECHO_RULES:
if rule[0] == family:
return rule
raise KeyError(family)
def matches_reasoning_echo_family(
family: str, provider: Any, model: Any, base_url: Any
) -> bool:
"""True when (provider, model, base_url) matches one echo-back family.
Families can overlap (e.g. a deepseek-named model pointed at a kimi
host); this membership test is independent per family so per-family
predicates keep their original semantics.
"""
from utils import base_url_host_matches
_, raw_providers, lowered_providers, model_subs, hosts = _family_rule(family)
provider_lower = (provider or "").lower()
model_lower = (model or "").lower()
if provider in raw_providers or provider_lower in lowered_providers:
return True
if any(sub in model_lower for sub in model_subs):
return True
return any(base_url_host_matches(base_url, host) for host in hosts)
def reasoning_echo_family(provider: Any, model: Any, base_url: Any) -> "str | None":
"""Classify the provider direction for the reasoning_content echo policy.
Returns ``"kimi"``, ``"deepseek"``, or ``"mimo"`` (first match in table
order) when the target endpoint enforces reasoning_content echo-back on
assistant turns, else ``None`` (strict/indifferent side — the field must
be stripped).
"""
for rule in _REASONING_ECHO_RULES:
if matches_reasoning_echo_family(rule[0], provider, model, base_url):
return rule[0]
return None
def needs_reasoning_echo(provider: Any, model: Any, base_url: Any) -> bool:
"""True when the endpoint requires reasoning_content echo-back."""
return reasoning_echo_family(provider, model, base_url) is not None
def apply_reasoning_content_policy(
source_msg: dict, api_msg: dict, needs_thinking_pad: bool
) -> None:
"""Copy provider-facing reasoning fields onto an API replay message.
``needs_thinking_pad`` is the require-side flag (see
``needs_reasoning_echo`` / the agent's cached
``_needs_thinking_reasoning_pad``). Mutates ``api_msg`` in place.
"""
if source_msg.get("role") != "assistant":
return
# 1. Explicit reasoning_content already set.
#
# When the active provider enforces the thinking-mode echo-back
# (DeepSeek / Kimi / MiMo), preserve it verbatim — that includes their
# own space-placeholder written at creation time and any valid reasoning
# from the same provider. Sessions persisted BEFORE #17341 have
# empty-string placeholders pinned at creation time; DeepSeek V4 Pro
# rejects those with HTTP 400, so upgrade "" → " " on replay.
#
# When the active provider does NOT enforce echo-back, strip the field
# entirely. Strict OpenAI-compatible providers (Mistral, Cerebras, Groq,
# SambaNova, …) reject ANY reasoning_content key in input messages with
# HTTP 400/422 ("Extra inputs are not permitted"), even an empty string
# or a single-space pad. This is the cross-provider fallback case: a
# reasoning primary (DeepSeek/Kimi/MiMo) pads history with " ", then a
# fallback to a strict provider replays that pad and 422s. Stripping
# here covers the rebuild path; ``reapply_reasoning_echo`` covers the
# already-built api_messages path. Refs #45655.
existing = source_msg.get("reasoning_content")
if isinstance(existing, str):
if not needs_thinking_pad:
api_msg.pop("reasoning_content", None)
elif existing == "":
api_msg["reasoning_content"] = " "
else:
api_msg["reasoning_content"] = existing
return
# 2. Cross-provider poisoned history (#15748): on DeepSeek/Kimi,
# if the source turn has tool_calls AND a 'reasoning' field but no
# 'reasoning_content' key, the 'reasoning' text was written by a
# prior provider (e.g. MiniMax) — DeepSeek's own _build_assistant_message
# pins reasoning_content at creation time for tool-call turns, so the
# shape (reasoning set, reasoning_content absent, tool_calls present)
# is unreachable from same-provider DeepSeek history after this fix.
# Inject a single space to satisfy the API without leaking another
# provider's chain of thought to DeepSeek/Kimi. Space (not "")
# because DeepSeek V4 Pro rejects empty-string reasoning_content
# in thinking mode (refs #17341).
normalized_reasoning = source_msg.get("reasoning")
if (
needs_thinking_pad
and source_msg.get("tool_calls")
and isinstance(normalized_reasoning, str)
and normalized_reasoning
):
api_msg["reasoning_content"] = " "
return
# 3. Healthy session: promote 'reasoning' field to 'reasoning_content'
# for providers that use the internal 'reasoning' key.
# This must happen before the unconditional empty-string fallback so
# genuine reasoning content is not overwritten (#15812 regression in
# PR #15478). Only promote for providers that enforce echo-back —
# strict providers reject the field (refs #45655).
if isinstance(normalized_reasoning, str) and normalized_reasoning:
if needs_thinking_pad:
api_msg["reasoning_content"] = normalized_reasoning
else:
api_msg.pop("reasoning_content", None)
return
# 4. DeepSeek / Kimi thinking mode: all assistant messages need
# reasoning_content. Inject a single space to satisfy the provider's
# requirement when no explicit reasoning content is present. Covers
# both tool-call turns (already-poisoned history with no reasoning
# at all) and plain text turns. Space (not "") because DeepSeek V4
# Pro tightened validation and rejects empty string with HTTP 400
# ("The reasoning content in the thinking mode must be passed back
# to the API"). Refs #17341.
if needs_thinking_pad:
api_msg["reasoning_content"] = " "
return
# 5. reasoning_content was present but not a string (e.g. None after
# context compaction). Don't pass null to the API.
api_msg.pop("reasoning_content", None)
def reapply_reasoning_echo(api_messages: list, needs_thinking_pad: bool) -> int:
"""Re-pad (or strip) assistant turns' reasoning_content for the active provider.
``api_messages`` is built once, before the retry loop, while the *primary*
provider is active. A mid-conversation fallback can then switch providers,
so the reasoning fields baked into ``api_messages`` are shaped for the
*prior* provider and must be reconciled against the *current* one:
* Switching TO a require-side provider (DeepSeek / Kimi / MiMo thinking
mode): assistant turns built when the prior provider did NOT need the
echo-back go out without ``reasoning_content`` and the new provider
rejects them with HTTP 400 ("The reasoning_content in the thinking mode
must be passed back"). Re-apply the pad.
* Switching TO a strict provider that rejects the field (Mistral,
Cerebras, Groq, SambaNova, …): assistant turns built under a reasoning
primary carry a ``reasoning_content`` pad (often a single space ``" "``),
and the strict provider rejects it with HTTP 400/422 ("Extra inputs are
not permitted"). Strip the field. This is the exact cross-provider
fallback bug from #45655 — a DeepSeek primary pads history with ``" "``,
the request falls back to Mistral, and Mistral 422s on the stale pad.
Calling this immediately before building the request kwargs reconciles the
fields against the *current* provider. It is idempotent and safe to call
every iteration; it covers every fallback path.
Returns the number of assistant turns whose reasoning_content was added or
removed.
"""
changed = 0
for api_msg in api_messages:
if api_msg.get("role") != "assistant":
continue
if needs_thinking_pad:
if api_msg.get("reasoning_content"):
continue
apply_reasoning_content_policy(api_msg, api_msg, needs_thinking_pad)
if api_msg.get("reasoning_content"):
changed += 1
else:
# Strict provider — strip any stale reasoning_content pad left
# over from a reasoning primary so the fallback request doesn't
# 400/422 on it.
if "reasoning_content" in api_msg:
api_msg.pop("reasoning_content", None)
changed += 1
return changed
# ---------------------------------------------------------------------------
# Image / multimodal parts — evaluated, NOT consolidated (verdict: syntax)
# ---------------------------------------------------------------------------
#
# The per-adapter image handling is format-specific SYNTAX, not shared policy:
# * anthropic_adapter (~1817): data-URL → Anthropic `source: {type: base64}`
# block mapping — Anthropic wire shape only.
# * codex_responses_adapter (~113/165/812): chat `image_url` parts →
# Responses `input_image` items and image counting for log summaries —
# Responses wire shape only.
# * transports/chat_completions: pass-through (native format).
# The one genuinely shared image POLICY — removing images when a server
# rejects them while preserving tool_call_id pairing — already has a single
# owner here: ``_strip_images_from_messages`` above.