surshar / claude-code

Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.

Built by Anthropic · github.com/anthropics

fix(hookify): add size limits to transcript reading to prevent OOMBrowse files

When hookify rules use `field: transcript`, the entire transcript file
was read into memory. For long-running sessions or ralph loops,
transcripts can grow to gigabytes, causing OOM kills.

This fix adds:
- 10MB max size limit for transcript reading
- For larger files, only the tail (most recent content) is read
- Warning at 5MB to alert users of large transcripts
- Proper error handling consolidated into a reusable function

This addresses memory issues reported in monorepo where CC was
consuming 30-40GB before getting OOM killed during startup/resume.

Claude authored on Jan 5, 2026 parent 5c92b97 · commit 449aa9d3b023

1 changed file+62 -16compared with 5c92b97

Mplugins/hookify/core/rule_engine.py

Mplugins/hookify/core/rule_engine.py+62 -16

		@@ -1,6 +1,7 @@
1	1	#!/usr/bin/env python3
2	2	"""Rule evaluation engine for hookify plugin."""
3	3
	4	+import os
4	5	import re
5	6	import sys
6	7	from functools import lru_cache
		@@ -9,7 +10,14 @@
9	10	# Import from local module
10	11	from hookify.core.config_loader import Rule, Condition
11	12
	13	+# Maximum transcript size to load into memory (10MB default)
	14	+# For larger files, only the tail is read to prevent OOM
	15	+MAX_TRANSCRIPT_SIZE_BYTES = 10 * 1024 * 1024 # 10MB
12	16
	17	+# Size threshold for warning about large transcripts
	18	+TRANSCRIPT_WARNING_SIZE_BYTES = 5 * 1024 * 1024 # 5MB
	19	+
	20	+
13	21	# Cache compiled regexes (max 128 patterns)
14	22	@lru_cache(maxsize=128)
15	23	def compile_regex(pattern: str) -> re.Pattern:
		@@ -24,6 +32,58 @@
24	32	return re.compile(pattern, re.IGNORECASE)
25	33
26	34
	35	+def read_transcript_safely(transcript_path: str) -> str:
	36	+ """Read transcript file with size limits to prevent OOM.
	37	+
	38	+ For large transcripts (>10MB), only reads the tail of the file
	39	+ to prevent memory exhaustion. This preserves the most recent
	40	+ conversation context which is typically what rules care about.
	41	+
	42	+ Args:
	43	+ transcript_path: Path to the transcript file
	44	+
	45	+ Returns:
	46	+ Transcript content as string, possibly truncated for large files
	47	+ """
	48	+ try:
	49	+ file_size = os.path.getsize(transcript_path)
	50	+
	51	+ # Warn about large transcripts
	52	+ if file_size > TRANSCRIPT_WARNING_SIZE_BYTES:
	53	+ size_mb = file_size / (1024 * 1024)
	54	+ print(f"Warning: Large transcript ({size_mb:.1f}MB): {transcript_path}", file=sys.stderr)
	55	+
	56	+ # For files within limit, read normally
	57	+ if file_size <= MAX_TRANSCRIPT_SIZE_BYTES:
	58	+ with open(transcript_path, 'r') as f:
	59	+ return f.read()
	60	+
	61	+ # For large files, read only the tail to prevent OOM
	62	+ size_mb = file_size / (1024 * 1024)
	63	+ limit_mb = MAX_TRANSCRIPT_SIZE_BYTES / (1024 * 1024)
	64	+ print(f"Warning: Transcript too large ({size_mb:.1f}MB), reading last {limit_mb:.0f}MB only", file=sys.stderr)
	65	+
	66	+ with open(transcript_path, 'r') as f:
	67	+ # Seek to position near end, leaving room for MAX_TRANSCRIPT_SIZE_BYTES
	68	+ f.seek(file_size - MAX_TRANSCRIPT_SIZE_BYTES)
	69	+ # Skip partial line at seek position
	70	+ f.readline()
	71	+ return f.read()
	72	+
	73	+ except FileNotFoundError:
	74	+ print(f"Warning: Transcript file not found: {transcript_path}", file=sys.stderr)
	75	+ return ''
	76	+ except PermissionError:
	77	+ print(f"Warning: Permission denied reading transcript: {transcript_path}", file=sys.stderr)
	78	+ return ''
	79	+ except (IOError, OSError) as e:
	80	+ print(f"Warning: Error reading transcript {transcript_path}: {e}", file=sys.stderr)
	81	+ return ''
	82	+ except UnicodeDecodeError as e:
	83	+ print(f"Warning: Encoding error in transcript {transcript_path}: {e}", file=sys.stderr)
	84	+ return ''
	85	+
	86	+
27	87	class RuleEngine:
28	88	"""Evaluates rules against hook input data."""
29	89
		@@ -205,24 +265,10 @@
205	265	if field == 'reason':
206	266	return input_data.get('reason', '')
207	267	elif field == 'transcript':
208		- # Read transcript file if path provided
	268	+ # Read transcript file with size limits to prevent OOM
209	269	transcript_path = input_data.get('transcript_path')
210	270	if transcript_path:
211		- try:
212		- with open(transcript_path, 'r') as f:
213		- return f.read()
214		- except FileNotFoundError:
215		- print(f"Warning: Transcript file not found: {transcript_path}", file=sys.stderr)
216		- return ''
217		- except PermissionError:
218		- print(f"Warning: Permission denied reading transcript: {transcript_path}", file=sys.stderr)
219		- return ''
220		- except (IOError, OSError) as e:
221		- print(f"Warning: Error reading transcript {transcript_path}: {e}", file=sys.stderr)
222		- return ''
223		- except UnicodeDecodeError as e:
224		- print(f"Warning: Encoding error in transcript {transcript_path}: {e}", file=sys.stderr)
225		- return ''
	271	+ return read_transcript_safely(transcript_path)
226	272	elif field == 'user_prompt':
227	273	# For UserPromptSubmit events
228	274	return input_data.get('user_prompt', '')