@ia-ccun/code-agent-cli 0.0.53 → 0.0.54-beta.3

package/scripts/regression/cases/03-rpc-protocol.sh

@@ -0,0 +1,259 @@
+#!/usr/bin/env bash
+# 03-rpc-protocol.sh
+# 验证：AicodeRpcClient.kt 实际用到的所有 RPC 命令 + 事件类型兼容性矩阵
+#
+# 这是 RPC 协议层的回归保护伞 — 任何 RPC 命令的字段格式变化都会在这里失败
+#
+# RPC 命令清单（来源：aicode-idea-plugin-swing/src/main/kotlin/com/nu/aicode/rpc/AicodeRpcClient.kt）
+#   状态控制:  prompt / steer / follow_up / abort / compact
+#   查询:      get_state / get_session_stats / get_commands
+#              get_available_models / get_available_thinking_levels
+#              get_messages / get_entries / get_tree
+#   模型控制:  set_model / cycle_model / set_model_provider
+#              set_thinking_level / cycle_thinking_level
+#   工具:      set_tools / bash
+#   会话:      new_session / switch_session / fork
+#   扩展钩子:  extension_ui_response
+#   新增 (0.84.4+): clear_queue
+#
+# 事件类型清单（同上文件 AicodeStreamItem sealed class）
+#   message_update(assistantMessageEvent: text_delta/thinking_delta/thinking)
+#   turn_start / turn_end / message_start / message_end
+#   agent_start / agent_end / agent_settled
+#   tool_execution_start / tool_execution_update / tool_execution_end
+#   queue_update / auto_retry_end
+
+set -uo pipefail
+CURRENT_CASE="03-rpc-protocol"
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+source "$SCRIPT_DIR/../lib/common.sh"
+
+header "$CURRENT_CASE" "RPC protocol compatibility matrix"
+
+if ! spawn_rpc 90; then
+    fail "spawn aicode RPC" "进程启动失败"
+    summary
+    exit 1
+fi
+pass "spawn aicode RPC (pid=$REGRESSION_PID)"
+
+# ----- Part 1: 所有 RPC 命令的兼容性 -----
+header "$CURRENT_CASE" "part 1: RPC command compatibility"
+
+# 每个命令: "type|payload|必需 response.success 字段|期望字段"
+# 简化版：只验证 type 字段能被识别（success=true 即视为兼容）
+# 失败类型：响应错误、字段缺失、命令不存在
+# RPC 命令兼容性测试 — 格式: "type|extra_fields_json_object"
+# extra_fields 是 type 之外的字段，会拼到 {"type":"X","id":"rpcN", extra_fields}
+#
+# 注意：只测 pi 核心协议命令。aicode-cli 自定义命令（set_model_provider、set_tools）
+# 是在 extensions 里实现的，跳过 -ne 时不生效，单独在 case 02 中测试。
+#
+# set_thinking_level / cycle_thinking_level 在并发发送场景下响应时序不稳定
+# （0.85.1 表现），交给 case 07 单独验证。
+RPC_COMMANDS=(
+    # 类型查询（pi 核心，必现）
+    "get_state|"
+    "get_session_stats|"
+    "get_commands|"
+    "get_available_models|"
+    "get_available_thinking_levels|"
+
+    # 会话管理
+    "new_session|"
+    "get_messages|"
+    "get_entries|"
+    "get_tree|"
+
+    # 新功能（0.84.4+）
+    "clear_queue|"
+)
+
+cmd_pass=0
+cmd_fail=0
+cmd_skip=0
+idx=0
+for spec in "${RPC_COMMANDS[@]}"; do
+    IFS='|' read -r cmd_type payload <<<"$spec"
+    idx=$((idx + 1))
+
+    req_id="rpc${idx}"
+    # 拼接合法 JSON：{"type":"X","id":"rpcN",<extra>}
+    if [[ -n "$payload" ]]; then
+        send "{\"type\":\"$cmd_type\",\"id\":\"$req_id\",$payload}"
+    else
+        send "{\"type\":\"$cmd_type\",\"id\":\"$req_id\"}"
+    fi
+
+    # 等最多 10 秒响应（set_tools/set_thinking_level 可能需要时间）
+    local_log="$REGRESSION_LOG"
+    waited=0
+    while [[ $waited -lt 20 ]]; do
+        if jq -e "select(.id==\"$req_id\" and .type==\"response\")" "$local_log" >/dev/null 2>&1; then
+            break
+        fi
+        sleep 0.5
+        waited=$((waited + 1))
+    done
+
+    resp=$(jq -c "select(.id==\"$req_id\" and .type==\"response\")" "$local_log" 2>/dev/null | head -1)
+    if [[ -z "$resp" ]]; then
+        fail "RPC $cmd_type 返回 response" "10s 内无响应"
+        cmd_fail=$((cmd_fail + 1))
+        continue
+    fi
+
+    success=$(echo "$resp" | jq -c '.success // false' 2>/dev/null)
+    if [[ "$success" == "true" ]]; then
+        pass "RPC $cmd_type success=true"
+        cmd_pass=$((cmd_pass + 1))
+    else
+        err=$(echo "$resp" | jq -c '.error // ""' 2>/dev/null)
+        # 区分"协议错误"和"业务错误"
+        if [[ -n "$err" ]] && echo "$err" | grep -qiE "Unknown command|not implemented|unsupported"; then
+            skip "RPC $cmd_type success=true" "pi 不支持: $err"
+            cmd_skip=$((cmd_skip + 1))
+        else
+            # 业务错误（如 fork 没 entryId）但协议 OK
+            if [[ "$success" == "false" ]] && [[ -n "$err" ]]; then
+                pass "RPC $cmd_type 协议 OK（业务错误: $err）"
+                cmd_pass=$((cmd_pass + 1))
+            else
+                fail "RPC $cmd_type 协议响应异常" "resp: $resp"
+                cmd_fail=$((cmd_fail + 1))
+            fi
+        fi
+    fi
+done
+
+echo ""
+echo "  RPC 命令小计: $cmd_pass pass / $cmd_fail fail / $cmd_skip skip"
+
+# ----- Part 2: 事件类型 + 字段格式（通过实际 prompt 触发）-----
+header "$CURRENT_CASE" "part 2: event type + field schema"
+
+# 发 prompt 触发完整事件流
+send '{"type":"prompt","message":"just say hi","id":"ev1"}'
+
+# 等 LLM 完成（以 agent_end 为信号，最长 120s）
+waited=0
+while [[ $waited -lt 120 ]]; do
+    # 检查两种结束信号：agent_end 或 agent_settled
+    if jq -e 'select(.type=="agent_end" or .type=="agent_settled")' "$REGRESSION_LOG" >/dev/null 2>&1; then
+        break
+    fi
+    sleep 1
+    waited=$((waited + 1))
+done
+
+if [[ $waited -ge 120 ]]; then
+    fail "LLM 在 120s 内完成" "超时"
+else
+    echo "    LLM 在 ${waited}s 内完成"
+fi
+
+# 事件类型清单（每个都要出现至少 1 次）
+EVENT_TYPES=(
+    "agent_start"
+    "turn_start"
+    "message_start"
+    "message_end"
+    "message_update"
+    "turn_end"
+    "agent_end"
+)
+
+ev_pass=0
+ev_fail=0
+for ev in "${EVENT_TYPES[@]}"; do
+    if jq -e "select(.type==\"$ev\")" "$REGRESSION_LOG" >/dev/null 2>&1; then
+        pass "事件 $ev 出现"
+        ev_pass=$((ev_pass + 1))
+    else
+        fail "事件 $ev 出现" "60s 内未观察到"
+        ev_fail=$((ev_fail + 1))
+    fi
+done
+
+echo ""
+echo "  事件类型小计: $ev_pass pass / $ev_fail fail"
+
+# ----- Part 3: 关键字段 schema 兼容性 -----
+header "$CURRENT_CASE" "part 3: event field schema"
+
+# message_start.message.role 必须是 "user" 或 "assistant"
+role_check=$(jq -c 'select(.type=="message_start") | .message.role // "missing"' "$REGRESSION_LOG" 2>/dev/null | sort -u | tr '\n' ' ')
+echo "    观察到的 roles: $role_check"
+if [[ "$role_check" == *"user"* && "$role_check" == *"assistant"* ]]; then
+    pass "message_start.message.role 包含 user + assistant"
+else
+    fail "message_start.message.role 字段" "got: $role_check"
+fi
+
+# turn_end.message.usage 字段
+usage_keys=$(jq -c 'select(.type=="turn_end") | .message.usage | keys | join(",")' "$REGRESSION_LOG" 2>/dev/null | head -1)
+echo "    turn_end.message.usage keys: $usage_keys"
+for key in input output cost; do
+    if [[ "$usage_keys" == *"$key"* ]]; then
+        pass "turn_end.usage.$key 存在"
+    else
+        fail "turn_end.usage.$key 存在" "got: $usage_keys"
+    fi
+done
+
+# message_update.assistantMessageEvent 子类型必须出现
+ame_subtypes=$(jq -r 'select(.type=="message_update") | .assistantMessageEvent.type // ""' "$REGRESSION_LOG" 2>/dev/null | sort -u | grep -v '^$' | tr '\n' ' ')
+echo "    assistantMessageEvent 子类型: $ame_subtypes"
+for subtype in thinking_delta text_delta; do
+    if [[ "$ame_subtypes" == *"$subtype"* ]]; then
+        pass "assistantMessageEvent.$subtype 出现"
+    else
+        fail "assistantMessageEvent.$subtype 出现" "got: $ame_subtypes"
+    fi
+done
+
+# message_update.assistantMessageEvent.delta 字段
+delta_check=$(jq -c 'select(.type=="message_update" and .assistantMessageEvent.type=="text_delta") | .assistantMessageEvent.delta // "missing"' "$REGRESSION_LOG" 2>/dev/null | head -1)
+if [[ "$delta_check" != '"missing"' && -n "$delta_check" ]]; then
+    pass "assistantMessageEvent.delta 字段存在"
+else
+    fail "assistantMessageEvent.delta 字段存在" "got: $delta_check"
+fi
+
+# turn_end 必须有 message 字段（不是 top-level usage）
+turn_end_schema=$(jq -c 'select(.type=="turn_end") | has("message") and (.message | has("usage"))' "$REGRESSION_LOG" 2>/dev/null | sort -u | head -1)
+if [[ "$turn_end_schema" == "true" ]]; then
+    pass "turn_end.message.usage 嵌套结构"
+else
+    fail "turn_end.message.usage 嵌套结构" "可能 schema 漂移：$turn_end_schema"
+fi
+
+# ----- Part 4: 计时关键字段（与 aicode-plugin-swing 状态计算相关）-----
+header "$CURRENT_CASE" "part 4: timing fields"
+
+# agent_end 可能含 willRetry 字段（aicode-plugin-swing 用它判断是否重试中）
+agent_end_keys=$(jq -c 'select(.type=="agent_end") | keys | join(",")' "$REGRESSION_LOG" 2>/dev/null | head -1)
+echo "    agent_end keys: $agent_end_keys"
+if [[ "$agent_end_keys" == *"willRetry"* ]]; then
+    pass "agent_end.willRetry 字段存在"
+else
+    skip "agent_end.willRetry 字段存在" "未出现（可能本轮不重试）"
+fi
+
+# message_end.stopReason 必须是已知值（0.85+ 新增 "stop" 通用值 + rawStopReason 保留 provider 原值）
+stop_reason=$(jq -c 'select(.type=="message_end" and .message.role=="assistant") | .message.stopReason // "missing"' "$REGRESSION_LOG" 2>/dev/null | sort -u | head -1)
+case "$stop_reason" in
+    '"end_turn"'|'"tool_use"'|'"error"'|'"max_tokens"'|'"stop_sequence"'|'"aborted"'|'"stop"')
+        pass "stopReason 合法 ($stop_reason)"
+        ;;
+    '"missing"')
+        fail "stopReason 字段" "未出现"
+        ;;
+    *)
+        skip "stopReason 已知值" "新值: $stop_reason（需人工确认是否要更新 AicodeRpcClient.kt）"
+        ;;
+esac
+
+cleanup_rpc
+summary
+exit $?

package/scripts/regression/cases/04-prompt-stream.sh

@@ -0,0 +1,130 @@
+#!/usr/bin/env bash
+# 03-prompt-stream.sh
+# 验证：prompt RPC 触发完整 message_update 事件流
+# 核心回归项 — AicodeRpcClient.kt 的 streamPrompt() 完全依赖这套事件序列
+#
+# 关键事件序列（与 aicode-plugin-swing AicodeStreamItem 映射）：
+#   1. {"type":"response","command":"prompt","success":true}   <- RPC 响应
+#   2. {"type":"agent_start"}
+#   3. {"type":"turn_start"}
+#   4. {"type":"message_start",message:{role:"user",...}}
+#   5. {"type":"message_end",message:{role:"user",...}}
+#   6. {"type":"message_start",message:{role:"assistant",api,provider,...}}
+#   7. N × {"type":"message_update",assistantMessageEvent:{type:"thinking_delta",delta:"..."}}
+#   8. {"type":"message_update",assistantMessageEvent:{type:"thinking_end",content:"..."}}
+#   9. {"type":"message_update",assistantMessageEvent:{type:"text_start",contentIndex:N}}
+#  10. N × {"type":"message_update",assistantMessageEvent:{type:"text_delta",delta:"..."}}
+#  11. {"type":"message_update",assistantMessageEvent:{type:"text_end",...}}  (可能)
+#  12. {"type":"message_end",message:{role:"assistant",stopReason:"end_turn"|"tool_use"|...}}
+#  13. {"type":"turn_end",message:{...usage:{input,output,cost,...}}}
+#  14. {"type":"agent_end"}
+
+set -uo pipefail
+CURRENT_CASE="03-prompt-stream"
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+source "$SCRIPT_DIR/../lib/common.sh"
+
+header "$CURRENT_CASE" "prompt → message_update stream"
+
+if ! spawn_rpc 60; then
+    fail "spawn aicode RPC" "进程启动失败"
+    summary
+    exit 1
+fi
+pass "spawn aicode RPC (pid=$REGRESSION_PID)"
+
+# 发 prompt
+send '{"type":"prompt","message":"reply with just: ok","id":"p1"}'
+
+# 等 LLM 完成（agent_end 作为信号，最长 120s）
+waited=0
+while [[ $waited -lt 120 ]]; do
+    if jq -e 'select(.type=="agent_end" or .type=="agent_settled")' "$REGRESSION_LOG" >/dev/null 2>&1; then
+        break
+    fi
+    sleep 0.5
+    waited=$((waited + 1))
+done
+
+if [[ $waited -ge 120 ]]; then
+    fail "LLM 在 120s 内完成" "超时"
+else
+    echo "    LLM 在 ${waited}.0s 内完成"
+fi
+
+# 验证事件序列
+header "$CURRENT_CASE" "event sequence"
+
+assert_event_count "agent_start 事件出现" \
+    'select(.type=="agent_start")' 1
+
+assert_event_count "turn_start 事件出现" \
+    'select(.type=="turn_start")' 1
+
+# user message_start / message_end 配对
+assert_event_count "user message_start 出现" \
+    'select(.type=="message_start" and .message.role=="user")' 1
+assert_event_count "user message_end 出现" \
+    'select(.type=="message_end" and .message.role=="user")' 1
+
+# assistant message_start
+assert_event_count "assistant message_start 出现" \
+    'select(.type=="message_start" and .message.role=="assistant")' 1
+
+# assistantMessageEvent 子类型
+header "$CURRENT_CASE" "assistantMessageEvent subtypes"
+
+# 至少 1 个 thinking_start
+assert_event_count "thinking_start 出现" \
+    'select(.type=="message_update" and .assistantMessageEvent.type=="thinking_start")' 1
+
+# 至少 1 个 thinking_delta (流式)
+assert_event_count "thinking_delta 至少 1 个" \
+    'select(.type=="message_update" and .assistantMessageEvent.type=="thinking_delta")' 1
+
+# 至少 1 个 text_start
+assert_event_count "text_start 出现" \
+    'select(.type=="message_update" and .assistantMessageEvent.type=="text_start")' 1
+
+# 至少 1 个 text_delta
+assert_event_count "text_delta 至少 1 个" \
+    'select(.type=="message_update" and .assistantMessageEvent.type=="text_delta")' 1
+
+# assistant message_end
+header "$CURRENT_CASE" "completion events"
+assert_event_count "assistant message_end 出现" \
+    'select(.type=="message_end" and .message.role=="assistant")' 1
+
+# stopReason 字段（end_turn / tool_use / error / stop 0.85+）
+stop_reason=$(jq -c 'select(.type=="message_end" and .message.role=="assistant") | .message.stopReason // "missing"' "$REGRESSION_LOG" 2>/dev/null | head -1)
+case "$stop_reason" in
+    '"end_turn"'|'"tool_use"'|'"error"'|'"max_tokens"'|'"stop_sequence"'|'"aborted"'|'"stop"')
+        pass "assistant stopReason 是 $stop_reason"
+        ;;
+    '"missing"')
+        fail "assistant stopReason 字段" "未出现"
+        ;;
+    *)
+        fail "assistant stopReason 合法" "未知值: $stop_reason"
+        ;;
+esac
+
+# turn_end 带 usage
+assert_event_count "turn_end 事件出现" \
+    'select(.type=="turn_end")' 1
+
+# usage 字段
+usage_input=$(jq -c 'select(.type=="turn_end") | .message.usage.input // "missing"' "$REGRESSION_LOG" 2>/dev/null | head -1)
+if [[ "$usage_input" != '"missing"' && "$usage_input" =~ ^[0-9]+$ ]]; then
+    pass "turn_end.usage.input 是数字 ($usage_input)"
+else
+    fail "turn_end.usage.input 是数字" "got: $usage_input"
+fi
+
+# agent_end 收尾
+assert_event_count "agent_end 事件出现" \
+    'select(.type=="agent_end")' 1
+
+cleanup_rpc
+summary
+exit $?

package/scripts/regression/cases/05-slash-commands-and-skills.sh

@@ -0,0 +1,119 @@
+#!/usr/bin/env bash
+# 04-slash-commands-and-skills.sh
+# 验证：aicode-cli 预置的 Prompts 与 Skills 全部能通过 get_commands 拿到
+#
+# Prompts (config/agent/prompts/):
+#   /code-simplifier /feat /git-commit /init-project /workflow
+# Skills (config/agent/skills/) — 通过 /skill: 触发:
+#   /skill:code-review /skill:commit /skill:pr /skill:refactor
+#   /skill:frontend-design /skill:brainstorming /skill:hello
+#   /skill:ponytail /skill:ponytail-audit /skill:ponytail-debt 等
+#
+# 这是 aicode-cli 的"开箱即用"卖点 — 升级 pi 后不能丢命令
+
+set -uo pipefail
+CURRENT_CASE="04-slash-commands-and-skills"
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+source "$SCRIPT_DIR/../lib/common.sh"
+
+header "$CURRENT_CASE" "prompts & skills registration"
+
+# pi 从 cwd 发现 prompts/skills/extensions。cd 到 aicode-cli 仓库根
+# 让 pi 能加载 config/agent/{prompts,skills,extensions}
+AICODE_CLI_ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"
+cd "$AICODE_CLI_ROOT" || { fail "cd aicode root" "$AICODE_CLI_ROOT"; summary; exit 1; }
+echo "    cwd: $AICODE_CLI_ROOT"
+
+if ! spawn_rpc 30 false; then
+    fail "spawn aicode RPC（带扩展）" "进程启动失败"
+    summary
+    exit 1
+fi
+pass "spawn aicode RPC（带扩展） (pid=$REGRESSION_PID)"
+
+# ----- Test 1: 总命令数 -----
+send '{"type":"get_commands","id":"c1"}'
+sleep 1
+
+cmds_json=$(jq -c 'select(.id=="c1" and .type=="response") | .data.commands' "$REGRESSION_LOG" 2>/dev/null | head -1)
+cmd_count=$(echo "$cmds_json" | jq -c 'length' 2>/dev/null)
+
+if [[ -z "$cmd_count" || "$cmd_count" == "null" ]]; then
+    fail "get_commands 返回 commands 数组" "got: $cmds_json"
+    cleanup_rpc; summary; exit 1
+fi
+echo "    total commands: $cmd_count"
+
+if [[ "$cmd_count" -lt 20 ]]; then
+    fail "至少注册 20 个命令（5 prompt + 20+ skill + 内置）" "got: $cmd_count"
+else
+    pass "注册 $cmd_count 个命令"
+fi
+
+# ----- Test 2: 5 个预置 Prompts -----
+header "$CURRENT_CASE" "predefined prompts (5)"
+
+PROMPTS=("git-commit" "workflow" "feat" "code-simplifier" "init-project")
+prompt_ok=0
+for p in "${PROMPTS[@]}"; do
+    # pi 0.85+ 的 commands.name 不带 / 前缀（如 "git-commit"），用户调用时才加 /
+    if echo "$cmds_json" | jq -e --arg c "$p" '.[] | select(.name == $c)' >/dev/null 2>&1; then
+        pass "Prompt /$p 注册"
+        prompt_ok=$((prompt_ok + 1))
+    else
+        fail "Prompt /$p 注册" "未在 commands 中"
+    fi
+done
+
+# ----- Test 3: 关键 Skills -----
+header "$CURRENT_CASE" "predefined skills (sample)"
+
+# 测试核心 skills（不强求全部，验证机制）
+SKILLS=("skill:code-review" "skill:commit" "skill:pr" "skill:refactor" "skill:frontend-design" "skill:brainstorming" "skill:hello")
+skill_ok=0
+for s in "${SKILLS[@]}"; do
+    # pi 0.85+ skills 用 "skill:xxx" 作为 name（含 skill: 前缀）
+    if echo "$cmds_json" | jq -e --arg c "$s" '.[] | select(.name == $c)' >/dev/null 2>&1; then
+        pass "Skill /$s 注册"
+        skill_ok=$((skill_ok + 1))
+    else
+        fail "Skill /$s 注册" "未在 commands 中"
+    fi
+done
+
+# 至少 60% 核心 skills 要在
+min_required=$(( ${#SKILLS[@]} * 60 / 100 ))
+if [[ $skill_ok -ge $min_required ]]; then
+    pass "核心 skills 注册率 ≥ 60% ($skill_ok/${#SKILLS[@]})"
+else
+    fail "核心 skills 注册率 ≥ 60%" "got: $skill_ok/${#SKILLS[@]}"
+fi
+
+# ----- Test 4: 每条 command 必须有 name + description -----
+header "$CURRENT_CASE" "command schema"
+
+# 抽 3 条 command 看字段
+sample=$(echo "$cmds_json" | jq -c '.[0:3]' 2>/dev/null)
+if [[ -n "$sample" ]]; then
+    echo "    sample commands:"
+    echo "$sample" | jq -r '.[] | "      - \(.name // "?") | \(.description // "(no description)")"' 2>/dev/null | head -3
+
+    # 每条都有 name
+    if echo "$cmds_json" | jq -e 'all(.[]; has("name"))' >/dev/null 2>&1; then
+        pass "所有 command 都有 name 字段"
+    else
+        fail "所有 command 都有 name 字段" "存在缺失 name 的 command"
+    fi
+fi
+
+# ----- Test 5: 内置命令（/model /settings /clear 等）也在 -----
+header "$CURRENT_CASE" "built-in commands"
+
+# RPC 模式下，pi 0.85.1 的内置命令（/model /settings 等）不通过 get_commands 返回
+# 它们是内置到 TUI 的，只在交互模式里出现。RPC 只返回 prompts/skills/extensions。
+# 跳过此检查，备注到 README。
+skip "内置命令 ≥ 5 个" "RPC 模式不返回内置命令（仅交互模式可见）"
+
+cleanup_rpc
+summary
+exit $?

package/scripts/regression/cases/06-tool-execution.sh

@@ -0,0 +1,115 @@
+#!/usr/bin/env bash
+# 05-tool-execution.sh
+# 验证：tool_execution_* 事件流（AicodeRpcClient.kt 的 ToolStart/ToolUpdate/ToolEnd 映射）
+#
+# 关键事件序列：
+#   1. {"type":"tool_execution_start",toolCallId,toolName,args}
+#   2. N × {"type":"tool_execution_update",toolCallId,partialResult:{content:[{text:"..."}]}}
+#   3. {"type":"tool_execution_end",toolCallId,result:{content:[{text:"..."}]},isError:"false"|"true"}
+#
+# 0.84.3 修复了 toolcall_start 缺 id/name 的 bug，所以 toolCallId 字段必须存在
+
+set -uo pipefail
+CURRENT_CASE="05-tool-execution"
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+source "$SCRIPT_DIR/../lib/common.sh"
+
+header "$CURRENT_CASE" "tool_execution event flow"
+
+if ! spawn_rpc 120; then
+    fail "spawn aicode RPC" "进程启动失败"
+    summary
+    exit 1
+fi
+pass "spawn aicode RPC (pid=$REGRESSION_PID)"
+
+# 触发 read：让模型读取一个已知文件
+TEST_FILE="/etc/hostname"
+send "{\"type\":\"prompt\",\"message\":\"use read tool to read /etc/hostname\",\"id\":\"t1\"}"
+
+# 等 turn 完成（最长 90s）
+header "$CURRENT_CASE" "waiting for turn"
+
+waited=0
+while [[ $waited -lt 90 ]]; do
+    if jq -e 'select(.id=="t1" and .type=="response")' "$REGRESSION_LOG" >/dev/null 2>&1; then
+        break
+    fi
+    sleep 1
+    waited=$((waited + 1))
+done
+
+if ! jq -e 'select(.id=="t1" and .type=="response")' "$REGRESSION_LOG" >/dev/null 2>&1; then
+    fail "prompt 在 90s 内完成" "超时"
+    cleanup_rpc; summary; exit 1
+fi
+
+# 验证 tool 事件
+header "$CURRENT_CASE" "tool_execution events"
+
+start_count=$(jq -c 'select(.type=="tool_execution_start")' "$REGRESSION_LOG" 2>/dev/null | wc -l | tr -d ' ')
+end_count=$(jq -c 'select(.type=="tool_execution_end")' "$REGRESSION_LOG" 2>/dev/null | wc -l | tr -d ' ')
+
+echo "    tool_execution_start: $start_count"
+echo "    tool_execution_end:   $end_count"
+
+if [[ $start_count -gt 0 ]]; then
+    pass "tool_execution_start 出现 ($start_count 个)"
+
+    # 0.84.3 修复：toolCallId 必现
+    tool_call_id=$(jq -c 'select(.type=="tool_execution_start") | .toolCallId // "missing"' "$REGRESSION_LOG" 2>/dev/null | head -1)
+    if [[ "$tool_call_id" != '"missing"' && "$tool_call_id" != '""' && -n "$tool_call_id" ]]; then
+        pass "tool_execution_start.toolCallId 存在 (0.84.3 修复点)"
+    else
+        fail "tool_execution_start.toolCallId 存在" "got: $tool_call_id"
+    fi
+
+    # 0.84.3 修复：toolName 必现
+    tool_name=$(jq -c 'select(.type=="tool_execution_start") | .toolName // "missing"' "$REGRESSION_LOG" 2>/dev/null | head -1)
+    if [[ "$tool_name" != '"missing"' && "$tool_name" != '""' && -n "$tool_name" ]]; then
+        pass "tool_execution_start.toolName 存在 (0.84.3 修复点) = $tool_name"
+    else
+        fail "tool_execution_start.toolName 存在" "got: $tool_name"
+    fi
+
+    # args 字段
+    args=$(jq -c 'select(.type=="tool_execution_start") | .args // "missing" | type' "$REGRESSION_LOG" 2>/dev/null | head -1)
+    if [[ "$args" == '"object"' ]]; then
+        pass "tool_execution_start.args 是 object"
+    else
+        fail "tool_execution_start.args 是 object" "got type: $args"
+    fi
+else
+    skip "tool_execution_start 出现" "模型没调用工具"
+fi
+
+if [[ $end_count -gt 0 ]]; then
+    pass "tool_execution_end 出现 ($end_count 个)"
+
+    # isError 字段
+    is_error=$(jq -c 'select(.type=="tool_execution_end") | .isError // "missing"' "$REGRESSION_LOG" 2>/dev/null | head -1)
+    if [[ "$is_error" == '"false"' || "$is_error" == '"true"' ]]; then
+        pass "tool_execution_end.isError 是 \"false\"|\"true\""
+    else
+        fail "tool_execution_end.isError 格式" "got: $is_error"
+    fi
+
+    # result.content 数组
+    result_type=$(jq -c 'select(.type=="tool_execution_end") | .result.content | type' "$REGRESSION_LOG" 2>/dev/null | head -1)
+    if [[ "$result_type" == '"array"' ]]; then
+        pass "tool_execution_end.result.content 是数组"
+    else
+        fail "tool_execution_end.result.content 是数组" "got: $result_type"
+    fi
+fi
+
+# start 与 end 配对
+if [[ $start_count -gt 0 && $start_count -eq $end_count ]]; then
+    pass "tool_execution_start 与 tool_execution_end 配对 ($start_count == $end_count)"
+elif [[ $start_count -gt 0 ]]; then
+    fail "tool_execution_start 与 tool_execution_end 配对" "$start_count start vs $end_count end"
+fi
+
+cleanup_rpc
+summary
+exit $?