Timmy_Foundation/hermes-agent
15
0
Fork 0
Code Issues 248 Pull Requests 14 Actions 1 Packages Projects Releases Wiki Activity

fix: improve read-loop detection — consecutive-only, correct thresholds, fix bugs

Browse Source 
Follow-up to PR #705 (merged from 0xbyt4). Addresses several issues:

1. CONSECUTIVE-ONLY TRACKING: Redesigned the read/search tracker to only
   warn/block on truly consecutive identical calls. Any other tool call
   in between (write, patch, terminal, etc.) resets the counter via
   notify_other_tool_call(), called from handle_function_call() in
   model_tools.py. This prevents false blocks in read→edit→verify flows.

2. THRESHOLD ADJUSTMENT: Warn on 3rd consecutive (was 2nd), block on
   4th+ consecutive (was 3rd+). Gives the model more room before
   intervening.

3. TUPLE UNPACKING BUG: Fixed get_read_files_summary() which crashed on
   search keys (5-tuple) when trying to unpack as 3-tuple. Now uses a
   separate read_history set that only tracks file reads.

4. WEB_EXTRACT DOCSTRING: Reverted incorrect removal of 'title' from
   web_extract return docs in code_execution_tool.py — the field IS
   returned by web_tools.py.

5. TESTS: Rewrote test_read_loop_detection.py (35 tests) to cover
   consecutive-only behavior, notify_other_tool_call, interleaved
   read/search, and summary-unaffected-by-searches.
This commit is contained in:
teknium1
2026-03-10 16:25:41 -07:00
parent b53d5dad67
commit a458b535c9
4 changed files with 223 additions and 55 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tools/code_execution_tool.py
View File

@@ -78,7 +78,7 @@ _TOOL_STUBS = {
"web_extract": (
"web_extract",
"urls: list",
'"""Extract content from URLs. Returns dict with results list of {url, content, error}."""',
'"""Extract content from URLs. Returns dict with results list of {url, title, content, error}."""',
'{"urls": urls}',
),
"read_file": (
@@ -616,7 +616,7 @@ _TOOL_DOC_LINES = [
" Returns {\"data\": {\"web\": [{\"url\", \"title\", \"description\"}, ...]}}"),
("web_extract",
" web_extract(urls: list[str]) -> dict\n"
" Returns {\"results\": [{\"url\", \"content\", \"error\"}, ...]} where content is markdown"),
" Returns {\"results\": [{\"url\", \"title\", \"content\", \"error\"}, ...]} where content is markdown"),
("read_file",
" read_file(path: str, offset: int = 1, limit: int = 500) -> dict\n"
" Lines are 1-indexed. Returns {\"content\": \"...\", \"total_lines\": N}"),

88
tools/file_tools.py
View File

@@ -15,7 +15,10 @@ _file_ops_lock = threading.Lock()
_file_ops_cache: dict = {}
# Track files read per task to detect re-read loops after context compression.
# Key: task_id, Value: dict mapping (path, offset, limit) -> read count
# Per task_id we store:
# "last_key": the key of the most recent read/search call (or None)
# "consecutive": how many times that exact call has been repeated in a row
# "read_history": set of (path, offset, limit) tuples for get_read_files_summary
_read_tracker_lock = threading.Lock()
_read_tracker: dict = {}
@@ -139,28 +142,37 @@ def read_file_tool(path: str, offset: int = 1, limit: int = 500, task_id: str =
result.content = redact_sensitive_text(result.content)
result_dict = result.to_dict()
# Track reads to detect re-read loops (e.g. after context compression)
read_key = (path, offset, limit)
# Track reads to detect *consecutive* re-read loops.
# The counter resets whenever any other tool is called in between,
# so only truly back-to-back identical reads trigger warnings/blocks.
read_key = ("read", path, offset, limit)
with _read_tracker_lock:
task_reads = _read_tracker.setdefault(task_id, {})
task_reads[read_key] = task_reads.get(read_key, 0) + 1
count = task_reads[read_key]
task_data = _read_tracker.setdefault(task_id, {
"last_key": None, "consecutive": 0, "read_history": set(),
})
task_data["read_history"].add((path, offset, limit))
if task_data["last_key"] == read_key:
task_data["consecutive"] += 1
else:
task_data["last_key"] = read_key
task_data["consecutive"] = 1
count = task_data["consecutive"]
if count >= 3:
if count >= 4:
# Hard block: stop returning content to break the loop
return json.dumps({
"error": (
f"BLOCKED: You have read this exact file region {count} times. "
f"BLOCKED: You have read this exact file region {count} times in a row. "
"The content has NOT changed. You already have this information. "
"STOP re-reading and proceed with your task."
),
"path": path,
"already_read": count,
}, ensure_ascii=False)
elif count > 1:
elif count >= 3:
result_dict["_warning"] = (
f"You have already read this exact file region {count} times in this session. "
"The content has not changed. Use the information you already have instead of re-reading. "
f"You have read this exact file region {count} times consecutively. "
"The content has not changed since your last read. Use the information you already have. "
"If you are stuck in a loop, stop reading and proceed with writing or responding."
)
@@ -176,9 +188,10 @@ def get_read_files_summary(task_id: str = "default") -> list:
compression boundaries.
"""
with _read_tracker_lock:
task_reads = _read_tracker.get(task_id, {})
seen_paths = {}
for (path, offset, limit), count in task_reads.items():
task_data = _read_tracker.get(task_id, {})
read_history = task_data.get("read_history", set())
seen_paths: dict = {}
for (path, offset, limit) in read_history:
if path not in seen_paths:
seen_paths[path] = []
seen_paths[path].append(f"lines {offset}-{offset + limit - 1}")
@@ -189,7 +202,12 @@ def get_read_files_summary(task_id: str = "default") -> list:
def clear_read_tracker(task_id: str = None):
"""Clear the read tracker. Called when starting a new conversation."""
"""Clear the read tracker.
Call with a task_id to clear just that task, or without to clear all.
Should be called when a session is destroyed to prevent memory leaks
in long-running gateway processes.
"""
with _read_tracker_lock:
if task_id:
_read_tracker.pop(task_id, None)
@@ -197,6 +215,22 @@ def clear_read_tracker(task_id: str = None):
_read_tracker.clear()
def notify_other_tool_call(task_id: str = "default"):
"""Reset consecutive read/search counter for a task.
Called by the tool dispatcher (model_tools.py) whenever a tool OTHER
than read_file / search_files is executed. This ensures we only warn
or block on *truly consecutive* repeated reads — if the agent does
anything else in between (write, patch, terminal, etc.) the counter
resets and the next read is treated as fresh.
"""
with _read_tracker_lock:
task_data = _read_tracker.get(task_id)
if task_data:
task_data["last_key"] = None
task_data["consecutive"] = 0
def write_file_tool(path: str, content: str, task_id: str = "default") -> str:
"""Write content to a file."""
try:
@@ -245,17 +279,23 @@ def search_tool(pattern: str, target: str = "content", path: str = ".",
task_id: str = "default") -> str:
"""Search for content or files."""
try:
# Track searches to detect repeated search loops
search_key = ("search", pattern, target, path, file_glob or "")
# Track searches to detect *consecutive* repeated search loops.
search_key = ("search", pattern, target, str(path), file_glob or "")
with _read_tracker_lock:
task_reads = _read_tracker.setdefault(task_id, {})
task_reads[search_key] = task_reads.get(search_key, 0) + 1
count = task_reads[search_key]
task_data = _read_tracker.setdefault(task_id, {
"last_key": None, "consecutive": 0, "read_history": set(),
})
if task_data["last_key"] == search_key:
task_data["consecutive"] += 1
else:
task_data["last_key"] = search_key
task_data["consecutive"] = 1
count = task_data["consecutive"]
if count >= 3:
if count >= 4:
return json.dumps({
"error": (
f"BLOCKED: You have run this exact search {count} times. "
f"BLOCKED: You have run this exact search {count} times in a row. "
"The results have NOT changed. You already have this information. "
"STOP re-searching and proceed with your task."
),
@@ -274,9 +314,9 @@ def search_tool(pattern: str, target: str = "content", path: str = ".",
m.content = redact_sensitive_text(m.content)
result_dict = result.to_dict()
if count > 1:
if count >= 3:
result_dict["_warning"] = (
f"You have run this exact search {count} times in this session. "
f"You have run this exact search {count} times consecutively. "
"The results have not changed. Use the information you already have."
)