diff --git a/agent/reviewer.py b/agent/reviewer.py
index ccf1841e..1bc232f8 100644
--- a/agent/reviewer.py
+++ b/agent/reviewer.py
@@ -47,7 +47,7 @@ from .middleware import (
     refresh_github_proxy_before_model,
     settle_review_check_on_exit,
 )
-from .reviewer_diff import compute_diff_line_set, fetch_pr_diff, fetch_pr_metadata
+from .reviewer_diff import compute_diff_line_set, fetch_pr_diff, fetch_pr_metadata, truncate_diff
 from .reviewer_findings import (
     list_findings as list_findings_async,
 )
@@ -847,9 +847,14 @@ async def get_reviewer_agent(config: RunnableConfig) -> Pregel:
         and bool(github_api_token)
     )
 
-    async def _fetch_diff_context() -> tuple[str, dict[str, dict[str, set[int]]] | None]:
+    async def _fetch_diff_context() -> tuple[str, str, dict[str, dict[str, set[int]]] | None]:
+        """Return (full_diff, truncated_diff, line_set).
+
+        The line set is computed from the full diff so findings on any changed
+        line are accepted. Only the prompt-facing text is truncated.
+        """
         if not can_fetch_pr or github_api_token is None or not isinstance(pr_number, int):
-            return "", None
+            return "", "", None
         fetched_diff = await fetch_pr_diff(
             owner=repo_owner,
             repo=repo_name,
@@ -857,8 +862,8 @@ async def get_reviewer_agent(config: RunnableConfig) -> Pregel:
             token=github_api_token,
         )
         if fetched_diff is None:
-            return "", None
-        return fetched_diff, compute_diff_line_set(fetched_diff)
+            return "", "", None
+        return fetched_diff, truncate_diff(fetched_diff), compute_diff_line_set(fetched_diff)
 
     async def _fetch_pr_overview() -> tuple[str, str]:
         if not can_fetch_pr or github_api_token is None or not isinstance(pr_number, int):
@@ -952,7 +957,7 @@ async def get_reviewer_agent(config: RunnableConfig) -> Pregel:
         _fetch_org_guidelines(),
         fetch_api_standards_skill(),
     )
-    pr_diff_text, pr_diff_line_set = diff_context
+    _, pr_diff_text, pr_diff_line_set = diff_context
     pr_title, pr_body = pr_overview
     config["configurable"]["diff_text"] = pr_diff_text
     config["configurable"]["diff_line_set"] = pr_diff_line_set
diff --git a/agent/reviewer_diff.py b/agent/reviewer_diff.py
index a8e1f0f9..ec6b68b8 100644
--- a/agent/reviewer_diff.py
+++ b/agent/reviewer_diff.py
@@ -197,6 +197,10 @@ def is_range_in_diff(
     return all(line in side_lines for line in range(start_line, end_line + 1))
 
 
+PR_DIFF_MAX_CHARS = 200_000
+PR_DIFF_TRUNCATION_MARKER = "\n... [PR diff truncated: {kept}/{total} chars]\n"
+
+
 async def fetch_pr_diff(
     *,
     owner: str,
@@ -211,6 +215,9 @@ async def fetch_pr_diff(
     validates against when posting inline review comments, so it's the right
     source for ``add_finding``'s in-diff anchor validation and for
     ``publish_review``'s 422 retry filter.
+
+    The full diff is returned — callers must use ``truncate_diff`` if they
+    need to cap the text before feeding it to the LLM.
     """
     import httpx
 
@@ -233,6 +240,19 @@ async def fetch_pr_diff(
     return response.text
 
 
+def truncate_diff(diff_text: str) -> str:
+    """Truncate diff text to ``PR_DIFF_MAX_CHARS`` with a visible marker.
+
+    Keeps the first half from the beginning and the second half from the end
+    so file headers and recent changes are both visible.
+    """
+    if len(diff_text) <= PR_DIFF_MAX_CHARS:
+        return diff_text
+    half = PR_DIFF_MAX_CHARS // 2
+    marker = PR_DIFF_TRUNCATION_MARKER.format(kept=PR_DIFF_MAX_CHARS, total=len(diff_text))
+    return diff_text[:half] + marker + diff_text[-half:]
+
+
 async def fetch_pr_metadata(
     *,
     owner: str,
diff --git a/agent/tools/fetch_url.py b/agent/tools/fetch_url.py
index 978a74da..aebc9afc 100644
--- a/agent/tools/fetch_url.py
+++ b/agent/tools/fetch_url.py
@@ -5,6 +5,8 @@ from markdownify import markdownify
 
 from .http_request import _request_with_safe_redirects
 
+FETCH_URL_MAX_CHARS = 100_000
+
 
 def fetch_url(url: str, timeout: int = 30) -> dict[str, Any]:
     """Fetch content from a URL and convert HTML to markdown format.
@@ -50,6 +52,12 @@ def fetch_url(url: str, timeout: int = 30) -> dict[str, Any]:
         # Convert HTML content to markdown
         markdown_content = markdownify(response.text)
 
+        if len(markdown_content) > FETCH_URL_MAX_CHARS:
+            markdown_content = (
+                markdown_content[:FETCH_URL_MAX_CHARS] + "\n... [content truncated: "
+                f"{FETCH_URL_MAX_CHARS}/{len(markdown_content)} chars]\n"
+            )
+
         return {
             "url": str(response.url),
             "markdown_content": markdown_content,
diff --git a/agent/tools/slack_read_thread_messages.py b/agent/tools/slack_read_thread_messages.py
index 25a16979..4b32eb86 100644
--- a/agent/tools/slack_read_thread_messages.py
+++ b/agent/tools/slack_read_thread_messages.py
@@ -2,6 +2,7 @@ import asyncio
 from typing import Any
 
 from ..utils.slack import (
+    SLACK_THREAD_MAX_MESSAGES,
     fetch_slack_thread_messages,
     format_slack_messages_for_prompt,
     get_slack_user_names,
@@ -19,8 +20,18 @@ async def _fetch_and_format(channel_id: str, message_ts: str) -> dict[str, Any]:
     ]
     user_names = await get_slack_user_names(user_ids) if user_ids else {}
 
+    truncated = len(messages) >= SLACK_THREAD_MAX_MESSAGES
     formatted = format_slack_messages_for_prompt(messages, user_names)
-    return {"success": True, "formatted": formatted, "count": len(messages)}
+    if truncated:
+        formatted = (
+            f"[thread truncated — showing most recent {len(messages)} messages]\n{formatted}"
+        )
+    return {
+        "success": True,
+        "formatted": formatted,
+        "count": len(messages),
+        "truncated": truncated,
+    }
 
 
 def slack_read_thread_messages(channel_id: str, message_ts: str) -> dict[str, Any]:
diff --git a/agent/utils/github_comments.py b/agent/utils/github_comments.py
index 7b771810..af85b996 100644
--- a/agent/utils/github_comments.py
+++ b/agent/utils/github_comments.py
@@ -44,6 +44,8 @@ _REACTION_ENDPOINTS: dict[str, str] = {
     "pull_request_review": "https://api.github.com/repos/{owner}/{repo}/pulls/{pull_number}/reviews/{comment_id}/reactions",
 }
 
+PAGINATED_MAX_PAGES = 50
+
 
 def verify_github_signature(body: bytes, signature: str, *, secret: str) -> bool:
     """Verify the GitHub webhook signature (X-Hub-Signature-256).
@@ -464,6 +466,9 @@ async def _fetch_paginated(
 ) -> list[dict[str, Any]]:
     """Fetch all pages from a GitHub paginated endpoint.
 
+    Caps at ``PAGINATED_MAX_PAGES`` pages to avoid unbounded fetching on
+    pathological PRs with thousands of comments.
+
     Args:
         client: An active httpx async client.
         url: The GitHub API endpoint URL.
@@ -475,7 +480,7 @@ async def _fetch_paginated(
     results: list[dict[str, Any]] = []
     params: dict[str, Any] = {"per_page": 100, "page": 1}
 
-    while True:
+    while params["page"] <= PAGINATED_MAX_PAGES:
         try:
             response = await client.get(url, headers=headers, params=params)
             if response.status_code == 401:
diff --git a/agent/utils/slack.py b/agent/utils/slack.py
index e260b42b..0f013812 100644
--- a/agent/utils/slack.py
+++ b/agent/utils/slack.py
@@ -24,6 +24,7 @@ logger = logging.getLogger(__name__)
 
 SLACK_API_BASE_URL = "https://slack.com/api"
 SLACK_BOT_TOKEN = os.environ.get("SLACK_BOT_TOKEN", "")
+SLACK_THREAD_MAX_MESSAGES = 500
 DEFAULT_ASSISTANT_STATUS = "is thinking…"
 
 # Curated rotating loading strings shown by Slack while the indicator is active.
@@ -504,12 +505,13 @@ async def get_slack_user_names(user_ids: list[str]) -> dict[str, str]:
 
 
 async def fetch_slack_thread_messages(channel_id: str, thread_ts: str) -> list[dict[str, Any]]:
-    """Fetch all messages for a Slack thread."""
+    """Fetch messages for a Slack thread, keeping the most recent window."""
     if not SLACK_BOT_TOKEN:
         return []
 
     messages: list[dict[str, Any]] = []
     cursor: str | None = None
+    truncated = False
 
     async with httpx.AsyncClient() as http_client:
         while True:
@@ -537,6 +539,16 @@ async def fetch_slack_thread_messages(channel_id: str, thread_ts: str) -> list[d
             if isinstance(batch, list):
                 messages.extend(item for item in batch if isinstance(item, dict))
 
+            if len(messages) >= SLACK_THREAD_MAX_MESSAGES:
+                truncated = True
+                logger.warning(
+                    "Slack thread %s/%s capped at %d messages",
+                    channel_id,
+                    thread_ts,
+                    SLACK_THREAD_MAX_MESSAGES,
+                )
+                break
+
             response_metadata = payload.get("response_metadata", {})
             cursor = (
                 response_metadata.get("next_cursor") if isinstance(response_metadata, dict) else ""
@@ -544,6 +556,8 @@ async def fetch_slack_thread_messages(channel_id: str, thread_ts: str) -> list[d
             if not cursor:
                 break
 
+    if truncated:
+        messages = messages[-SLACK_THREAD_MAX_MESSAGES:]
     messages.sort(key=lambda item: _parse_ts(item.get("ts")))
     return messages
 
diff --git a/agent/utils/thread_ops.py b/agent/utils/thread_ops.py
index 447d7915..fb38d837 100644
--- a/agent/utils/thread_ops.py
+++ b/agent/utils/thread_ops.py
@@ -10,6 +10,8 @@ from langgraph_sdk import get_client
 
 logger = logging.getLogger(__name__)
 
+MAX_QUEUED_MESSAGES = 100
+
 
 def langgraph_url() -> str:
     return os.environ.get("LANGGRAPH_URL") or os.environ.get(
@@ -57,6 +59,13 @@ async def queue_message_for_thread(
             logger.debug("No existing queued messages for thread %s", thread_id)
 
         existing_messages.append(new_message)
+        if len(existing_messages) > MAX_QUEUED_MESSAGES:
+            existing_messages = existing_messages[-MAX_QUEUED_MESSAGES:]
+            logger.warning(
+                "Thread %s queue capped at %d messages (dropped oldest)",
+                thread_id,
+                MAX_QUEUED_MESSAGES,
+            )
         await client.store.put_item(namespace, key, {"messages": existing_messages})
         logger.info(
             "Queued message for thread %s (total queued: %d)",