Skip to content

fix: feed back ExitPlanMode approval so CLI doesn't loop the plan #65

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
quick-sort merged 1 commit into from
Jun 19, 2026
Merged

fix: feed back ExitPlanMode approval so CLI doesn't loop the plan #65

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
189 changes: 145 additions & 44 deletions src/agent_box/agents/claude_code.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,7 +34,16 @@
# we forward the question to the user via the IM channel and pause the
# generator until the user replies. The next ``run()`` call sends the
# reply back as a ``tool_result`` so the CLI can continue.
_TOOLS_REQUIRING_USER_INPUT = frozenset({"AskUserQuestion"})
#
# ExitPlanMode is included because Claude Code's CLI treats it as
# ``requiresUserInteraction()`` — it blocks waiting for an approval
# tool_result. Without feeding one back, the next user message would
# re-enter the blocked CLI and the plan would be re-emitted in a loop.
_TOOLS_REQUIRING_USER_INPUT = frozenset({"AskUserQuestion", "ExitPlanMode"})

# Hint appended to surfaced plans so the user knows how to respond.
# Reply "Yes" to approve, or "No <feedback>" to send feedback to the agent.
_PLAN_APPROVAL_HINT = "\n\n— 回复 Yes 批准 / No <修改意见> 退回修改"

# Regex to match [SEND_FILE:/path/to/file] markers in agent text output.
_SEND_FILE_RE = re.compile(r"\[SEND_FILE:([^\]]+)\]")
Expand Down Expand Up @@ -327,6 +336,60 @@ def _format_answers_for_tool_result(answers: dict) -> str:
)


# ── ExitPlanMode approval parser ──

# Words that count as approval / rejection when the user replies to a plan.
# Matched as a prefix on the reply (case-insensitive) so Chinese phrases
# without word boundaries (不要用 / 不行,) work just as well as English
# ("Yes, please", "No, change step 2"). Anything that doesn't match either
# defaults to approval so the conversation doesn't stall — the plan is
# already visible to the user.
_PLAN_APPROVE_WORDS = (
"yes", "y", "ok", "okay", "approve", "approved", "lgtm",
"是", "好的", "好", "可以", "同意", "批准", "行",
)
_PLAN_REJECT_WORDS = (
"no", "n", "reject", "rejected", "deny", "denied", "cancel",
"否", "不要", "不行", "拒绝", "别", "不",
)


def _build_plan_approval_result(user_reply: str) -> str:
"""Translate the user's Yes/No reply into the tool_result for ExitPlanMode.

- Approval word prefix → "User has approved your plan..."
- Rejection word prefix + following text → rejection with the rest as feedback
- Anything else → rejection with the reply as feedback (don't run a plan
the user didn't explicitly approve)
"""
reply = (user_reply or "").strip()
reply_lower = reply.lower()

# Reject: longest prefix match wins (so "不要用" matches "不要" before "不")
reject_match = max(
(w for w in _PLAN_REJECT_WORDS if reply_lower.startswith(w.lower())),
key=len, default=None,
)
if reject_match:
feedback = reply[len(reject_match):].lstrip(" :：,，、").strip()
if feedback:
return (
f"User rejected the plan. Feedback: {feedback}. "
"Please revise the plan based on this feedback and try again."
)
return "User rejected the plan. Please revise and try again."

if any(reply_lower.startswith(w.lower()) for w in _PLAN_APPROVE_WORDS):
return "User has approved your plan. You can now start coding."

# Ambiguous — default to rejection so the agent doesn't run with a plan
# the user didn't explicitly approve. The reply is attached as feedback.
return (
f"User rejected the plan. Feedback: {reply}. "
"Please revise the plan based on this feedback and try again."
)


class ClaudeCodeAgent(BaseAgent):
"""Each project gets one ClaudeSDKClient. Session id is tracked externally."""

Expand Down Expand Up @@ -495,31 +558,38 @@ def has_pending_question(self) -> bool:
async def run(self, prompt: str, user_id: str = "", channel: str = "") -> AsyncIterator[OutgoingMessage]:
client = await self._ensure_client()

# If a previous run() paused with a pending AskUserQuestion, the
# current prompt is the user's answer — send it as tool_result.
# If a previous run() paused with a pending AskUserQuestion or
# ExitPlanMode, the current prompt is the user's answer — send it
# as a tool_result so the blocked CLI can continue.
if self._pending_ask is not None:
ask = self._pending_ask
self._pending_ask = None
questions = ask.get("questions", [])
kind = ask.get("kind", "question")
log.info(
"resuming pending AskUserQuestion tool_use_id=%s",
ask["tool_use_id"],
"resuming pending %s tool_use_id=%s",
kind, ask["tool_use_id"],
)
try:
parsed = await _parse_user_answer(questions, prompt)
content = _format_answers_for_tool_result(parsed["answers"])
log.info("parsed AskUserQuestion answers: %s", parsed["answers"])
except Exception:
log.warning(
"failed to parse AskUserQuestion answer with LLM, "
"falling back to raw user reply",
exc_info=True,
)
content = (
"User has answered your questions: "
f"{prompt}. You can now continue with the user's "
"answers in mind."
)

if kind == "exit_plan_mode":
content = _build_plan_approval_result(prompt)
log.info("ExitPlanMode reply parsed: %s", content)
else:
questions = ask.get("questions", [])
try:
parsed = await _parse_user_answer(questions, prompt)
content = _format_answers_for_tool_result(parsed["answers"])
log.info("parsed AskUserQuestion answers: %s", parsed["answers"])
except Exception:
log.warning(
"failed to parse AskUserQuestion answer with LLM, "
"falling back to raw user reply",
exc_info=True,
)
content = (
"User has answered your questions: "
f"{prompt}. You can now continue with the user's "
"answers in mind."
)
await self._send_tool_result(
client,
tool_use_id=ask["tool_use_id"],
Expand All @@ -537,12 +607,41 @@ async def run(self, prompt: str, user_id: str = "", channel: str = "") -> AsyncI
async for msg in client.receive_response():
if isinstance(msg, AssistantMessage):
for block in msg.content:
# --- AskUserQuestion interception ---
# --- Tools that require user interaction ---
# Both AskUserQuestion and ExitPlanMode block the CLI
# waiting for a tool_result, so we surface their content
# to the IM channel, set _pending_ask, and return. The
# next run() with the user's reply feeds back the
# tool_result so the CLI can continue.
if (
isinstance(block, ToolUseBlock)
and block.name in _TOOLS_REQUIRING_USER_INPUT
):
# Log the full input for debugging
if block.name == "ExitPlanMode":
plan = (block.input or {}).get("plan")
body = plan.strip() if isinstance(plan, str) else ""
if not body:
body = "(agent 请求退出 plan 模式，但没有提供计划内容)"
text = body + _PLAN_APPROVAL_HINT
log.info(
"ExitPlanMode detected, tool_use_id=%s",
block.id,
)
yield OutgoingMessage(
text=text,
user_id=user_id,
channel=channel,
type=MessageType.text,
data={"id": block.id, "name": block.name, "input": block.input},
)
self._pending_ask = {
"tool_use_id": block.id,
"session_id": msg.session_id,
"kind": "exit_plan_mode",
}
return # Stop yielding; next run() will resume

# AskUserQuestion
log.info(
"AskUserQuestion detected, tool_use_id=%s, input=%s",
block.id,
Expand All @@ -559,6 +658,7 @@ async def run(self, prompt: str, user_id: str = "", channel: str = "") -> AsyncI
"tool_use_id": block.id,
"session_id": msg.session_id,
"questions": block.input.get("questions", []) if block.input else [],
"kind": "question",
}
log.info(
"AskUserQuestion intercepted, pausing run() "
Expand All @@ -582,18 +682,6 @@ async def run(self, prompt: str, user_id: str = "", channel: str = "") -> AsyncI
data={"file_path": fp},
)
elif isinstance(block, ToolUseBlock):
# ExitPlanMode carries the plan text in its input — surface
# the full plan so the user can review it on the IM channel,
# instead of the generic "⚙️ ExitPlanMode" one-liner.
if block.name == "ExitPlanMode":
plan = (block.input or {}).get("plan")
if isinstance(plan, str) and plan.strip():
yield OutgoingMessage(
text=plan.strip(), user_id=user_id, channel=channel,
type=MessageType.text,
data={"id": block.id, "name": block.name, "input": block.input},
)
continue
# Brief one-liner so the user knows something is happening.
summary = _format_tool_summary(block, prefixes=_path_prefixes)
yield OutgoingMessage(
Expand Down Expand Up @@ -689,6 +777,28 @@ async def _recover_from_context_limit(
async for msg in client.receive_response():
if isinstance(msg, AssistantMessage):
for block in msg.content:
# ExitPlanMode blocks the CLI waiting for approval —
# surface plan + hint, set pending state, and pause so
# the next run() can feed back the user's Yes/No reply.
if (
isinstance(block, ToolUseBlock)
and block.name == "ExitPlanMode"
):
plan = (block.input or {}).get("plan")
body = plan.strip() if isinstance(plan, str) else ""
if not body:
body = "(agent 请求退出 plan 模式，但没有提供计划内容)"
yield OutgoingMessage(
text=body + _PLAN_APPROVAL_HINT,
user_id=user_id, channel=channel, type=MessageType.text,
data={"id": block.id, "name": block.name, "input": block.input},
)
self._pending_ask = {
"tool_use_id": block.id,
"session_id": msg.session_id,
"kind": "exit_plan_mode",
}
return
if isinstance(block, TextBlock):
cleaned = block.text.strip()
if cleaned:
Expand All @@ -701,15 +811,6 @@ async def _recover_from_context_limit(
data={"file_path": fp},
)
elif isinstance(block, ToolUseBlock):
if block.name == "ExitPlanMode":
plan = (block.input or {}).get("plan")
if isinstance(plan, str) and plan.strip():
yield OutgoingMessage(
text=plan.strip(), user_id=user_id, channel=channel,
type=MessageType.text,
data={"id": block.id, "name": block.name, "input": block.input},
)
continue
summary = _format_tool_summary(block, prefixes=_path_prefixes)
yield OutgoingMessage(
text=summary, user_id=user_id, channel=channel, type=MessageType.text,
Expand Down
Loading
Loading