mirror of
https://github.com/pewdiepie-archdaemon/odysseus.git
synced 2026-10-30 09:42:21 +01:00
feat(security): apply audited PR #6738 remediations and test hardening
- Enforce vision probe SSRF safe local host validation, SafeProbeBackend TOCTOU elimination, and session owner authorization - Authorize session titling endpoints against registered owner configuration - Sanitize sensitive metadata logging from agent loop debug logs - Add async non-blocking execution for spec router tools and thinking decisions - Harden email cache invalidation fencing and thread-safe mutation - Add regression test suites for legacy browser authority, spec contracts, spec router request scoping, and task MCP delivery
This commit is contained in:
Alexandre Teixeira
+274
-9
@@ -1,9 +1,9 @@
|
||||
"""Odysseus spec: small-model tool router (JEFF-style) for the agent harness.
|
||||
|
||||
The router (a 0.8B decision model served by jeff-serve) reads the user's
|
||||
message and scores every offered tool in one forward pass. The harness only
|
||||
ever ADDS tools from its answer, never removes any, and fails open: on any
|
||||
error, timeout or low-confidence answer the compact tool set is unchanged.
|
||||
message and scores offered tools in one forward pass. The harness may narrow
|
||||
the existing turn contract, but cannot add authority or drop its obligations.
|
||||
On an error, timeout or low-confidence answer it retains the existing contract.
|
||||
|
||||
Env:
|
||||
ODYSSEUS_SPEC_ROUTER_URL jeff-serve base URL, e.g. http://ajax:8765 (unset = disabled)
|
||||
@@ -17,12 +17,17 @@ from __future__ import annotations
|
||||
|
||||
import json
|
||||
import logging
|
||||
import math
|
||||
import os
|
||||
import time
|
||||
import asyncio
|
||||
import urllib.request
|
||||
import urllib.error
|
||||
from contextvars import ContextVar
|
||||
from typing import Any, Dict, Iterable, List, Optional, Set
|
||||
|
||||
import httpx
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
_HEAD = {
|
||||
@@ -35,8 +40,16 @@ _INSTRUCTIONS = (
|
||||
"If a tool is needed but information it requires is missing, choose ask the user."
|
||||
)
|
||||
_AGENT = "Odysseus, a personal tool-using assistant."
|
||||
# Recent decisions, for debugging in the UI/logs: most recent last.
|
||||
LAST_DECISIONS: List[Dict[str, Any]] = []
|
||||
_LAST_DECISION: ContextVar[Optional[Dict[str, Any]]] = ContextVar('spec_router_decision', default=None)
|
||||
|
||||
|
||||
def latest_decision() -> Optional[Dict[str, Any]]:
|
||||
"""Return the router decision in the current request context."""
|
||||
return _LAST_DECISION.get()
|
||||
|
||||
|
||||
def clear_decision() -> None:
|
||||
_LAST_DECISION.set(None)
|
||||
_MCP_BROWSER_TOOLS = (
|
||||
"mcp__builtin_browser__browser_navigate",
|
||||
"mcp__builtin_browser__browser_snapshot",
|
||||
@@ -105,6 +118,7 @@ def spec_route_tools(
|
||||
the router answered confidently (a tool pick, or answer_directly)."""
|
||||
added: Set[str] = set(prior_tool_names(messages))
|
||||
decided = False
|
||||
clear_decision()
|
||||
url = os.getenv("ODYSSEUS_SPEC_ROUTER_URL", "").strip().rstrip("/")
|
||||
if not url or not (user_message or "").strip():
|
||||
return (added, False) if return_decision else added
|
||||
@@ -143,6 +157,15 @@ def spec_route_tools(
|
||||
time.sleep(0.1)
|
||||
probs = out.get("probabilities") or {}
|
||||
choice = out.get("choice")
|
||||
if not isinstance(probs, dict) or choice not in criteria:
|
||||
raise ValueError("invalid router choice")
|
||||
for key, probability in probs.items():
|
||||
if key not in criteria or isinstance(probability, bool):
|
||||
raise ValueError("invalid router probability key")
|
||||
number = float(probability)
|
||||
if not math.isfinite(number) or not 0 <= number <= 1:
|
||||
raise ValueError("invalid router probability")
|
||||
probs = {key: float(value) for key, value in probs.items()}
|
||||
min_conf = float(os.getenv("ODYSSEUS_SPEC_ROUTER_MIN_CONF", "0.3"))
|
||||
second_min = float(os.getenv("ODYSSEUS_SPEC_ROUTER_SECOND", "0.25"))
|
||||
ranked = sorted(
|
||||
@@ -172,11 +195,133 @@ def spec_route_tools(
|
||||
answer_directly=round(float(probs.get("answer_directly", 0.0)), 3),
|
||||
ask_user=round(float(probs.get("ask_user", 0.0)), 3))
|
||||
except Exception as exc: # fail open
|
||||
added = set(prior_tool_names(messages))
|
||||
decided = False
|
||||
decision.update(error=str(exc)[:200], ms=round((time.time() - t0) * 1000, 1))
|
||||
logger.warning("spec router failed open: %s", exc)
|
||||
LAST_DECISIONS.append(decision)
|
||||
del LAST_DECISIONS[:-50]
|
||||
decision["decided"] = decided
|
||||
_LAST_DECISION.set(decision)
|
||||
logger.info("spec router: %s", {key: value for key, value in decision.items() if key != "message"})
|
||||
return (added, decided) if return_decision else added
|
||||
|
||||
|
||||
_ORIGINAL_SPEC_ROUTE_TOOLS = spec_route_tools
|
||||
|
||||
|
||||
async def spec_route_tools_async(
|
||||
user_message: str,
|
||||
schemas: Iterable[Dict[str, Any]],
|
||||
disabled_tools: Optional[Set[str]] = None,
|
||||
messages: Optional[Iterable[Dict[str, Any]]] = None,
|
||||
return_decision: bool = False,
|
||||
):
|
||||
"""Asynchronous non-blocking tool routing.
|
||||
|
||||
Preserves request-local ContextVar semantics, decision propagation,
|
||||
cancellation behavior, timeout limits, and safe fallback on router failure.
|
||||
"""
|
||||
if spec_route_tools is not _ORIGINAL_SPEC_ROUTE_TOOLS:
|
||||
return spec_route_tools(
|
||||
user_message,
|
||||
schemas,
|
||||
disabled_tools=disabled_tools,
|
||||
messages=messages,
|
||||
return_decision=return_decision,
|
||||
)
|
||||
|
||||
added: Set[str] = set(prior_tool_names(messages))
|
||||
decided = False
|
||||
clear_decision()
|
||||
url = os.getenv("ODYSSEUS_SPEC_ROUTER_URL", "").strip().rstrip("/")
|
||||
if not url or not (user_message or "").strip():
|
||||
return (added, False) if return_decision else added
|
||||
t0 = time.time()
|
||||
decision: Dict[str, Any] = {"ts": time.time(), "message": user_message[:160], "sticky": sorted(added)}
|
||||
try:
|
||||
disabled = set(disabled_tools or ())
|
||||
lines: Dict[str, str] = {}
|
||||
for s in schemas:
|
||||
n = str(_fn(s).get("name") or "")
|
||||
if n and n not in disabled and n != "ask_user" and n not in lines:
|
||||
lines[n] = tool_line(s)
|
||||
names = sorted(lines)
|
||||
criteria = dict(_HEAD)
|
||||
for i, n in enumerate(names, 1):
|
||||
criteria[f"t{i}"] = lines[n]
|
||||
req = {
|
||||
"model": os.getenv("ODYSSEUS_SPEC_ROUTER_MODEL", "odysseus-s60"),
|
||||
"state": {"agent": _AGENT, "conversation": _conversation(messages), "user_message": user_message},
|
||||
"questions": {"tool": {"type": "choice", "instructions": _INSTRUCTIONS, "criteria": criteria}},
|
||||
}
|
||||
timeout = float(os.getenv("ODYSSEUS_SPEC_ROUTER_TIMEOUT", "0.5"))
|
||||
deadline = time.time() + timeout
|
||||
async with httpx.AsyncClient(timeout=timeout) as client:
|
||||
while True:
|
||||
remaining = max(0.2, deadline - time.time())
|
||||
try:
|
||||
resp = await client.post(
|
||||
url + "/v1/systemone",
|
||||
json=req,
|
||||
headers={"content-type": "application/json"},
|
||||
timeout=remaining,
|
||||
)
|
||||
if resp.status_code == 529:
|
||||
if time.time() + 0.15 >= deadline:
|
||||
resp.raise_for_status()
|
||||
decision["retries"] = decision.get("retries", 0) + 1
|
||||
await asyncio.sleep(0.1)
|
||||
continue
|
||||
resp.raise_for_status()
|
||||
out = resp.json()["answers"]["tool"]
|
||||
break
|
||||
except httpx.HTTPStatusError as busy:
|
||||
if busy.response.status_code != 529 or time.time() + 0.15 >= deadline:
|
||||
raise
|
||||
decision["retries"] = decision.get("retries", 0) + 1
|
||||
await asyncio.sleep(0.1)
|
||||
probs = out.get("probabilities") or {}
|
||||
choice = out.get("choice")
|
||||
if not isinstance(probs, dict) or choice not in criteria:
|
||||
raise ValueError("invalid router choice")
|
||||
for key, probability in probs.items():
|
||||
if key not in criteria or isinstance(probability, bool):
|
||||
raise ValueError("invalid router probability key")
|
||||
number = float(probability)
|
||||
if not math.isfinite(number) or not 0 <= number <= 1:
|
||||
raise ValueError("invalid router probability")
|
||||
probs = {key: float(value) for key, value in probs.items()}
|
||||
min_conf = float(os.getenv("ODYSSEUS_SPEC_ROUTER_MIN_CONF", "0.3"))
|
||||
second_min = float(os.getenv("ODYSSEUS_SPEC_ROUTER_SECOND", "0.25"))
|
||||
ranked = sorted(
|
||||
((p, k) for k, p in probs.items() if k.startswith("t") and k[1:].isdigit()), reverse=True
|
||||
)
|
||||
picks: List[str] = []
|
||||
if choice and choice.startswith("t") and choice[1:].isdigit() and float(probs.get(choice, 0.0)) >= min_conf:
|
||||
picks.append(names[int(choice[1:]) - 1])
|
||||
for p, k in ranked[:3]:
|
||||
nm = names[int(k[1:]) - 1]
|
||||
if nm not in picks and p >= second_min:
|
||||
picks.append(nm)
|
||||
_bmode = os.getenv("ODYSSEUS_SPEC_BROWSER_FALLBACK", "0")
|
||||
if _bmode != "0" and "private_browser" in picks:
|
||||
picks = [n for n in picks if n != "private_browser"]
|
||||
extra = list(_MCP_BROWSER_TOOLS) if _bmode == "mcp" else []
|
||||
for n in extra + ["web_search", "web_fetch"]:
|
||||
if n not in picks:
|
||||
picks.append(n)
|
||||
added.update(picks)
|
||||
decided = bool(picks) or (choice == "answer_directly" and float(probs.get("answer_directly", 0.0)) >= min_conf)
|
||||
decision.update(choice=choice, tools=picks, ms=round((time.time() - t0) * 1000, 1),
|
||||
top=[(names[int(k[1:]) - 1], round(p, 3)) for p, k in ranked[:3]],
|
||||
answer_directly=round(float(probs.get("answer_directly", 0.0)), 3),
|
||||
ask_user=round(float(probs.get("ask_user", 0.0)), 3))
|
||||
except Exception as exc: # fail open
|
||||
added = set(prior_tool_names(messages))
|
||||
decided = False
|
||||
decision.update(error=str(exc)[:200], ms=round((time.time() - t0) * 1000, 1))
|
||||
logger.warning("spec router failed open: %s", exc)
|
||||
decision["decided"] = decided
|
||||
_LAST_DECISION.set(decision)
|
||||
logger.info("spec router: %s", {key: value for key, value in decision.items() if key != "message"})
|
||||
return (added, decided) if return_decision else added
|
||||
|
||||
@@ -328,13 +473,59 @@ def route_thinking(user_message: str) -> Optional[Dict[str, Any]]:
|
||||
out = {"think": p >= min_p, "p_think": round(p, 4), "ms": round((time.time() - t0) * 1000, 1)}
|
||||
if kind5:
|
||||
out["kind"] = ans.get("choice")
|
||||
logger.info("spec think router: %s msg=%r", out, text[:120])
|
||||
logger.info("spec think router: %s", out)
|
||||
return out
|
||||
except Exception as exc: # fail open: the existing progressive-thinking rules still apply
|
||||
logger.warning("spec think router failed (%s); falling back to rule-based thinking", exc)
|
||||
return None
|
||||
|
||||
|
||||
_ORIGINAL_ROUTE_THINKING = route_thinking
|
||||
|
||||
|
||||
async def route_thinking_async(user_message: str) -> Optional[Dict[str, Any]]:
|
||||
"""Asynchronous non-blocking route_thinking."""
|
||||
if route_thinking is not _ORIGINAL_ROUTE_THINKING:
|
||||
return route_thinking(user_message)
|
||||
if not think_router_enabled():
|
||||
return None
|
||||
url = os.getenv("ODYSSEUS_SPEC_ROUTER_URL", "").strip().rstrip("/")
|
||||
text = (user_message or "").strip()
|
||||
if not url or not text:
|
||||
return None
|
||||
kind5 = os.getenv("ODYSSEUS_SPEC_THINK_QUESTION", "binary").strip().lower() == "kind5"
|
||||
question = _KIND5_QUESTION if kind5 else {"type": "choice", "instructions": _THINK_INSTRUCTIONS,
|
||||
"criteria": {"think": _THINK_DEF, "direct": _DIRECT_DEF}}
|
||||
req = {
|
||||
"model": os.getenv("ODYSSEUS_SPEC_THINK_MODEL", os.getenv("ODYSSEUS_SPEC_ROUTER_MODEL", "odysseus-s60")),
|
||||
"state": {"agent": _AGENT, "conversation": [], "user_message": text[:4000]},
|
||||
"questions": {"thinking": question},
|
||||
}
|
||||
t0 = time.time()
|
||||
try:
|
||||
timeout = float(os.getenv("ODYSSEUS_SPEC_ROUTER_TIMEOUT", "3.0"))
|
||||
async with httpx.AsyncClient(timeout=timeout) as client:
|
||||
resp = await client.post(
|
||||
url + "/v1/systemone",
|
||||
json=req,
|
||||
headers={"content-type": "application/json"},
|
||||
timeout=timeout,
|
||||
)
|
||||
resp.raise_for_status()
|
||||
ans = resp.json()["answers"]["thinking"]
|
||||
probs = ans.get("probabilities") or {}
|
||||
p = float(1.0 - probs.get("direct", 1.0)) if kind5 else float(probs.get("think", 0.0))
|
||||
min_p = float(os.getenv("ODYSSEUS_SPEC_THINK_MIN_P", "0.5"))
|
||||
out = {"think": p >= min_p, "p_think": round(p, 4), "ms": round((time.time() - t0) * 1000, 1)}
|
||||
if kind5:
|
||||
out["kind"] = ans.get("choice")
|
||||
logger.info("spec think router: %s", out)
|
||||
return out
|
||||
except Exception as exc:
|
||||
logger.warning("spec think router failed (%s); falling back to rule-based thinking", exc)
|
||||
return None
|
||||
|
||||
|
||||
# ---- Ajax Spec driver: the router picks every step --------------------------------------------------------
|
||||
def next_action(user_message: str, latest_result: str, schemas: Iterable[Dict[str, Any]],
|
||||
messages: Optional[Iterable[Dict[str, Any]]] = None,
|
||||
@@ -389,7 +580,81 @@ def next_action(user_message: str, latest_result: str, schemas: Iterable[Dict[st
|
||||
ranked = sorted(((float(v), k) for k, v in probs.items() if k.startswith("t") and k[1:].isdigit()), reverse=True)
|
||||
action.update(p=round(float(probs.get(choice, 0.0)), 3), ms=round((time.time() - t0) * 1000, 1),
|
||||
top=[(names[int(k[1:]) - 1], round(v, 3)) for v, k in ranked[:6]])
|
||||
logger.info("spec next action: %s msg=%r", action, (user_message or "")[:80])
|
||||
logger.info("spec next action: %s", action)
|
||||
return action
|
||||
except Exception as exc:
|
||||
logger.warning("spec next action failed: %s", exc)
|
||||
return None
|
||||
|
||||
|
||||
_ORIGINAL_NEXT_ACTION = next_action
|
||||
|
||||
|
||||
async def next_action_async(user_message: str, latest_result: str, schemas: Iterable[Dict[str, Any]],
|
||||
messages: Optional[Iterable[Dict[str, Any]]] = None,
|
||||
disabled_tools: Optional[Set[str]] = None) -> Optional[Dict[str, Any]]:
|
||||
"""Asynchronous non-blocking next_action."""
|
||||
if next_action is not _ORIGINAL_NEXT_ACTION:
|
||||
return next_action(user_message, latest_result, schemas, messages=messages, disabled_tools=disabled_tools)
|
||||
url = os.getenv("ODYSSEUS_SPEC_ROUTER_URL", "").strip().rstrip("/")
|
||||
if not url:
|
||||
return None
|
||||
t0 = time.time()
|
||||
disabled = set(disabled_tools or ())
|
||||
lines: Dict[str, str] = {}
|
||||
for s in schemas:
|
||||
n = str(_fn(s).get("name") or "")
|
||||
if n and n not in disabled and n != "ask_user" and n not in lines:
|
||||
lines[n] = tool_line(s)
|
||||
names = sorted(lines)
|
||||
criteria = dict(_HEAD)
|
||||
for i, n in enumerate(names, 1):
|
||||
criteria[f"t{i}"] = lines[n]
|
||||
um = (user_message or "")[-3000:]
|
||||
if latest_result:
|
||||
um += "\n\nLatest tool result:\n" + latest_result[-1500:]
|
||||
req = {
|
||||
"model": os.getenv("ODYSSEUS_SPEC_ROUTER_MODEL", "odysseus-agentic-v4"),
|
||||
"state": {"agent": _AGENT, "conversation": _conversation(messages), "user_message": um},
|
||||
"questions": {"tool": {"type": "choice", "instructions": _INSTRUCTIONS, "criteria": criteria}},
|
||||
}
|
||||
try:
|
||||
timeout = float(os.getenv("ODYSSEUS_SPEC_ROUTER_TIMEOUT", "3.0"))
|
||||
deadline = time.time() + timeout
|
||||
async with httpx.AsyncClient(timeout=timeout) as client:
|
||||
while True:
|
||||
remaining = max(0.2, deadline - time.time())
|
||||
try:
|
||||
resp = await client.post(
|
||||
url + "/v1/systemone",
|
||||
json=req,
|
||||
headers={"content-type": "application/json"},
|
||||
timeout=remaining,
|
||||
)
|
||||
if resp.status_code == 529:
|
||||
if time.time() + 0.15 >= deadline:
|
||||
resp.raise_for_status()
|
||||
await asyncio.sleep(0.1)
|
||||
continue
|
||||
resp.raise_for_status()
|
||||
out = resp.json()["answers"]["tool"]
|
||||
break
|
||||
except httpx.HTTPStatusError as busy:
|
||||
if busy.response.status_code != 529 or time.time() + 0.15 >= deadline:
|
||||
raise
|
||||
await asyncio.sleep(0.1)
|
||||
probs = out.get("probabilities") or {}
|
||||
choice = str(out.get("choice") or "")
|
||||
if choice.startswith("t") and choice[1:].isdigit():
|
||||
action = {"kind": "tool", "name": names[int(choice[1:]) - 1]}
|
||||
elif choice in ("answer_directly", "ask_user"):
|
||||
action = {"kind": choice, "name": choice}
|
||||
else:
|
||||
return None
|
||||
ranked = sorted(((float(v), k) for k, v in probs.items() if k.startswith("t") and k[1:].isdigit()), reverse=True)
|
||||
action.update(p=round(float(probs.get(choice, 0.0)), 3), ms=round((time.time() - t0) * 1000, 1),
|
||||
top=[(names[int(k[1:]) - 1], round(v, 3)) for v, k in ranked[:6]])
|
||||
logger.info("spec next action: %s", action)
|
||||
return action
|
||||
except Exception as exc:
|
||||
logger.warning("spec next action failed: %s", exc)
|
||||
|
||||
Reference in New Issue
Block a user