mirror of
https://github.com/pewdiepie-archdaemon/odysseus.git
synced 2026-07-09 12:07:18 +00:00
544a664766
Review follow-up on #3681 (thanks @vgalin): 1. Routing bare email names made 10 of the 14 email tools executable by non-admin owners — is_public_blocked_tool() runs on the bare name before dispatch, and NON_ADMIN_BLOCKED_TOOLS only listed 4. Define the full email tool set once (BUILTIN_EMAIL_TOOLS in tool_security.py) and derive the blocklist, the fence tags (TOOL_TAGS), the bare-name dispatch, and the native-call mapping from it so they can't drift. This also fixes 4 tools (search_emails, draft_email, draft_email_reply, ai_draft_email_reply) that were missing from the old tool_schemas copy and therefore unreachable even for native function-calling models. 2. The relaxed fence regex from the previous commit could prefix-match longer fence tags: ```python3 parsed as tool "python" with content "3\nprint(...)" and executed as code. Add a (?![\w-]) boundary after the tag. Tests: test_public_agent_policy_blocks_sensitive_tools now covers all 14 bare email names + the mcp__email__ form; new tests/test_fenced_inline_args.py pins inline-args parsing, the python3/hyphenated-tag non-matches, and strip/parse display mirroring. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
72 lines
3.0 KiB
Python
72 lines
3.0 KiB
Python
"""PR #3681 — fenced tool calls with inline args, and the fence-tag boundary.
|
|
|
|
Local fenced-block models (Ollama etc.) emit calls like ```list_email_accounts {}
|
|
with the args on the same line as the tag; the parser must execute those. The
|
|
relaxed tag pattern must NOT prefix-match longer fence tags: ```python3 is a
|
|
language hint, not a "python" tool call with content "3\n...".
|
|
"""
|
|
import sys
|
|
from unittest.mock import MagicMock
|
|
|
|
for mod in ['src.agent_tools', 'src.tool_parsing', 'src.tool_schemas', 'src.tool_execution']:
|
|
sys.modules.pop(mod, None)
|
|
for mod in [
|
|
'sqlalchemy', 'sqlalchemy.orm', 'sqlalchemy.ext', 'sqlalchemy.ext.declarative',
|
|
'sqlalchemy.ext.hybrid', 'sqlalchemy.sql', 'sqlalchemy.sql.expression',
|
|
'src.database', 'core.models', 'core.database', 'core.auth'
|
|
]:
|
|
if mod not in sys.modules:
|
|
sys.modules[mod] = MagicMock()
|
|
|
|
import src.agent_tools # noqa: E402, F401
|
|
from src.tool_parsing import parse_tool_blocks, strip_tool_blocks # noqa: E402
|
|
|
|
|
|
def test_inline_args_on_tag_line_parse():
|
|
# The original bug: ```list_email_accounts {} (args on the tag line)
|
|
# never matched because the regex required a newline right after the tag.
|
|
blocks = parse_tool_blocks('```list_email_accounts {}\n```')
|
|
assert [(b.tool_type, b.content) for b in blocks] == [("list_email_accounts", "{}")]
|
|
|
|
|
|
def test_inline_json_args_parse_for_email_tools():
|
|
blocks = parse_tool_blocks('```list_emails {"max_results": 5}\n```')
|
|
assert [(b.tool_type, b.content) for b in blocks] == [("list_emails", '{"max_results": 5}')]
|
|
|
|
|
|
def test_next_line_content_still_parses():
|
|
# No regression for the classic shape: tag, newline, content.
|
|
blocks = parse_tool_blocks('```manage_memory\nadd\nsome text\n```')
|
|
assert [(b.tool_type, b.content) for b in blocks] == [("manage_memory", "add\nsome text")]
|
|
|
|
|
|
def test_plain_bash_fence_still_parses():
|
|
blocks = parse_tool_blocks('```bash\necho hello\n```')
|
|
assert [(b.tool_type, b.content) for b in blocks] == [("bash", "echo hello")]
|
|
|
|
|
|
def test_python3_language_hint_is_not_a_python_tool_call():
|
|
# ```python3 must not prefix-match the "python" fence tag — without the
|
|
# (?![\w-]) boundary it parsed as tool "python" with content "3\nprint(...)"
|
|
# and executed as code.
|
|
blocks = parse_tool_blocks('```python3\nprint("hi")\n```')
|
|
assert blocks == [], blocks
|
|
|
|
|
|
def test_hyphenated_tag_is_not_a_tool_call():
|
|
blocks = parse_tool_blocks('```bash-session\n$ ls\n```')
|
|
assert blocks == [], blocks
|
|
|
|
|
|
def test_inline_args_fence_is_stripped_from_display():
|
|
# strip must mirror parse: an executed inline-args fence must not leak
|
|
# into the displayed text.
|
|
text = 'Checking now.\n```list_email_accounts {}\n```\nDone.'
|
|
assert strip_tool_blocks(text) == 'Checking now.\n\nDone.'
|
|
|
|
|
|
def test_python3_fence_is_left_intact_in_display():
|
|
# ...and a fence that did NOT parse as a tool call must stay visible.
|
|
text = 'Example:\n```python3\nprint("hi")\n```'
|
|
assert strip_tool_blocks(text) == text
|