@ia-ccun/code-agent-cli 0.0.52 → 0.0.53-beta.1

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 $?

package/scripts/regression/cases/07-pi-new-features.sh

@@ -0,0 +1,141 @@
+#!/usr/bin/env bash
+# 06-pi-new-features.sh
+# 验证：pi 0.85+ 新功能在 aicode 端可用
+#
+# 重点：
+#   - clear_queue RPC（0.84.4+ 新增）
+#   - GPT-6 Astra 是否出现在 available_models
+#   - thinking 持久化（set_thinking_level 后 turn 间不丢）
+#   - usage.cacheWrite1h 字段（0.85+ 新增）
+#   - 持久化 Claude thinking effort（0.85+）
+#   - /thinking 命令选择器（0.84.3+）
+
+set -uo pipefail
+CURRENT_CASE="06-pi-new-features"
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+source "$SCRIPT_DIR/../lib/common.sh"
+
+header "$CURRENT_CASE" "pi 0.85+ new features"
+
+if ! spawn_rpc 90; then
+    fail "spawn aicode RPC" "进程启动失败"
+    summary
+    exit 1
+fi
+pass "spawn aicode RPC (pid=$REGRESSION_PID)"
+
+# ----- Test 1: clear_queue RPC -----
+header "$CURRENT_CASE" "clear_queue RPC (0.84.4+)"
+
+send '{"type":"clear_queue","id":"n1"}'
+sleep 0.5
+
+if jq -e 'select(.id=="n1" and .type=="response" and .success==true)' "$REGRESSION_LOG" >/dev/null 2>&1; then
+    pass "clear_queue RPC 成功"
+
+    # 验证返回结构
+    if jq -e 'select(.id=="n1" and .type=="response") | .data | has("steering") and has("followUp")' "$REGRESSION_LOG" >/dev/null 2>&1; then
+        pass "clear_queue 返回 {steering,followUp} 字段"
+    else
+        skip "clear_queue 返回 {steering,followUp} 字段" "字段名可能不同（followUps/follow_ups）"
+    fi
+else
+    err=$(jq -c 'select(.id=="n1" and .type=="response") | .error // ""' "$REGRESSION_LOG" 2>/dev/null | head -1)
+    if [[ -n "$err" && "$err" != '""' ]]; then
+        skip "clear_queue RPC" "pi < 0.84.4 不支持: $err"
+    else
+        fail "clear_queue RPC" "无 response"
+    fi
+fi
+
+# ----- Test 2: GPT-6 Astra 在 available_models -----
+header "$CURRENT_CASE" "GPT-6 Astra (0.85.0+)"
+
+send '{"type":"get_available_models","id":"n2"}'
+sleep 1
+
+models=$(jq -c 'select(.id=="n2" and .type=="response") | .data' "$REGRESSION_LOG" 2>/dev/null | head -1)
+if [[ -n "$models" ]]; then
+    if echo "$models" | jq -e '.[] | select(.id | tostring | contains("gpt-6"))' >/dev/null 2>&1; then
+        pass "GPT-6 系列模型在 catalog 中"
+        gpt6=$(echo "$models" | jq -r '.[] | select(.id | tostring | contains("gpt-6")) | .id' 2>/dev/null | head -3)
+        echo "      发现: $(echo "$gpt6" | tr '\n' ' ')"
+    else
+        skip "GPT-6 Astra 在 catalog 中" "未找到 gpt-6* 模型（取决于 provider 配置）"
+    fi
+else
+    fail "get_available_models 返回 data" "无响应"
+fi
+
+# ----- Test 3: set_thinking_level + 持久化（0.85.0）-----
+header "$CURRENT_CASE" "thinking level persistence (0.85.0)"
+
+# 设置 high
+send '{"type":"set_thinking_level","level":"high","id":"n3a"}'
+sleep 0.5
+if jq -e 'select(.id=="n3a" and .type=="response" and .success==true)' "$REGRESSION_LOG" >/dev/null 2>&1; then
+    pass "set_thinking_level high 成功"
+else
+    fail "set_thinking_level high" "无响应"
+fi
+
+# 再发一个 prompt 让它跑
+send '{"type":"prompt","message":"say hi","id":"n3b"}'
+sleep 8
+
+# 等 prompt 完成
+waited=0
+while [[ $waited -lt 60 ]]; do
+    if jq -e 'select(.id=="n3b" and .type=="response")' "$REGRESSION_LOG" >/dev/null 2>&1; then
+        break
+    fi
+    sleep 1
+    waited=$((waited + 1))
+done
+
+# 查询 get_state — thinking level 应该还是 high
+send '{"type":"get_state","id":"n3c"}'
+sleep 0.5
+
+state_level=$(jq -c 'select(.id=="n3c" and .type=="response") | .data.thinkingLevel // "missing"' "$REGRESSION_LOG" 2>/dev/null | head -1)
+if [[ "$state_level" == '"high"' ]]; then
+    pass "thinking level 在 turn 间保持 high（持久化）"
+else
+    if [[ "$state_level" == '"medium"' ]]; then
+        skip "thinking level 持久化" "回到 medium（0.85 之前行为）"
+    else
+        fail "thinking level 持久化" "got: $state_level"
+    fi
+fi
+
+# ----- Test 4: usage.cacheWrite1h 字段（0.85.0+）-----
+header "$CURRENT_CASE" "usage.cacheWrite1h field (0.85.0+)"
+
+# 看 turn_end 事件的 usage 是否包含 cacheWrite1h
+has_cw1h=$(jq -c 'select(.type=="turn_end") | .message.usage | has("cacheWrite1h")' "$REGRESSION_LOG" 2>/dev/null | head -1)
+if [[ "$has_cw1h" == "true" ]]; then
+    pass "turn_end.usage.cacheWrite1h 字段存在"
+else
+    skip "turn_end.usage.cacheWrite1h 字段" "不存在（pi < 0.85.0 或 provider 不报告）"
+fi
+
+# ----- Test 5: get_available_thinking_levels -----
+header "$CURRENT_CASE" "thinking levels list"
+
+send '{"type":"get_available_thinking_levels","id":"n5"}'
+sleep 0.5
+
+if jq -e 'select(.id=="n5" and .type=="response" and .success==true)' "$REGRESSION_LOG" >/dev/null 2>&1; then
+    levels=$(jq -c 'select(.id=="n5" and .type=="response") | .data | length' "$REGRESSION_LOG" 2>/dev/null | head -1)
+    if [[ -n "$levels" && "$levels" -gt 0 ]]; then
+        pass "get_available_thinking_levels 返回 $levels 个级别"
+    else
+        fail "thinking levels 非空" "got: $levels"
+    fi
+else
+    fail "get_available_thinking_levels" "无响应"
+fi
+
+cleanup_rpc
+summary
+exit $?

package/scripts/regression/lib/common.sh

@@ -0,0 +1,287 @@
+#!/usr/bin/env bash
+# common.sh - 回归测试公共工具
+# 用法：在 case 脚本开头 source "$(dirname "${BASH_SOURCE[0]}")/lib/common.sh"
+
+# 严格模式：未定义变量报错，但允许管道部分失败（grep 找不到时 exit 1）
+set -uo pipefail
+
+# 默认配置
+: "${AICODE_BIN:=aicode}"                    # aicode 二进制路径
+: "${AICODE_CONFIG:=$HOME/.aicode-cli}"      # 配置目录
+: "${AICODE_TIMEOUT:=30}"                    # RPC 测试默认超时（秒）
+: "${REGRESSION_QUIET:=0}"                   # 1 = 减少输出
+
+# 路径
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+RESULTS_DIR="$SCRIPT_DIR/results"
+mkdir -p "$RESULTS_DIR"
+
+# 颜色（仅 TTY 启用）
+if [[ -t 1 ]] && [[ "${REGRESSION_QUIET:-0}" != "1" ]]; then
+    RED=$'\033[0;31m'; GREEN=$'\033[0;32m'; YELLOW=$'\033[0;33m'; BLUE=$'\033[0;34m'; CYAN=$'\033[0;36m'; DIM=$'\033[2m'; NC=$'\033[0m'
+else
+    RED=''; GREEN=''; YELLOW=''; BLUE=''; CYAN=''; DIM=''; NC=''
+fi
+
+# 计数器
+PASS_COUNT=0
+FAIL_COUNT=0
+SKIP_COUNT=0
+CURRENT_CASE="${CURRENT_CASE:-$(basename "${BASH_SOURCE[1]:-unknown}" .sh)}"
+
+# ----- 输出 -----
+header() {
+    [[ "${REGRESSION_QUIET:-0}" == "1" ]] && return
+    echo ""
+    echo "${CYAN}─── $CURRENT_CASE: $2 ───${NC}"
+}
+
+pass() {
+    local name="$1"
+    echo "  ${GREEN}✓${NC} $name"
+    PASS_COUNT=$((PASS_COUNT + 1))
+}
+
+fail() {
+    local name="$1"
+    local reason="${2:-no reason}"
+    echo "  ${RED}✗${NC} $name"
+    echo "    ${DIM}└─ $reason${NC}"
+    FAIL_COUNT=$((FAIL_COUNT + 1))
+}
+
+skip() {
+    local name="$1"
+    local reason="${2:-no reason}"
+    echo "  ${YELLOW}○${NC} $name ${DIM}(skipped: $reason)${NC}"
+    SKIP_COUNT=$((SKIP_COUNT + 1))
+}
+
+summary() {
+    local exit_code=0
+    echo ""
+    if [[ $FAIL_COUNT -eq 0 && $SKIP_COUNT -eq 0 ]]; then
+        echo "${GREEN}✓ $CURRENT_CASE: ALL PASS${NC}  ${DIM}($PASS_COUNT checks)${NC}"
+    elif [[ $FAIL_COUNT -eq 0 ]]; then
+        echo "${YELLOW}△ $CURRENT_CASE: PASS with skips${NC}  ${DIM}($PASS_COUNT pass, $SKIP_COUNT skip)${NC}"
+    else
+        echo "${RED}✗ $CURRENT_CASE: FAILED${NC}  ${DIM}($FAIL_COUNT fail, $PASS_COUNT pass, $SKIP_COUNT skip)${NC}"
+        exit_code=1
+    fi
+    return $exit_code
+}
+
+# ----- RPC 进程管理 -----
+#
+# spawn_rpc [timeout_seconds]
+#   启动 aicode -ne --mode rpc --no-session，stdin 从 fd 11 读，stdout/stderr 到 fd 12 写
+#   用法见 cases/02-rpc-handshake.sh
+#
+spawn_rpc() {
+    local timeout="${1:-$AICODE_TIMEOUT}"
+    local skip_extensions="${2:-true}"
+
+    if ! command -v "$AICODE_BIN" >/dev/null 2>&1; then
+        echo "ERROR: $AICODE_BIN not found. Set AICODE_BIN env var to override." >&2
+        return 127
+    fi
+
+    # 临时配置必须在 $HOME 下，aicode-cli 的 config-loader.ts 安全检查
+    # 会拒绝非 home 目录并回退到 ~/.aicode-cli
+    # 用 home 下的隐藏目录避免污染用户配置
+    local workdir="$HOME/.aicode-regression-$$-$(date +%s%N | tail -c 8)"
+    mkdir -p "$workdir/agent/extensions"
+    local cfg="$workdir/config.json"
+    cat >"$cfg" <<EOF
+{
+  "name": "aicode-regression",
+  "banner": "",
+  "themeColor": "#49bccf",
+  "configDir": "$(basename "$workdir")"
+}
+EOF
+
+    # 复用用户全局的 auth.json / models.json / settings.json + prompts / skills / extensions
+    # （避免重新登录/配置，且能让 pi 加载预置 prompts 和 skills）
+    # ~/.pi 通常是 ~/.aicode-cli 的 symlink
+    local user_agent_dir="$HOME/.aicode-cli/agent"
+    if [[ -d "$user_agent_dir" ]]; then
+        for f in auth.json models.json settings.json; do
+            [[ -f "$user_agent_dir/$f" ]] && cp "$user_agent_dir/$f" "$workdir/agent/$f" 2>/dev/null
+        done
+        # 复制 prompts/skills/extensions 目录（不覆盖本地）
+        for d in prompts skills extensions themes; do
+            if [[ -d "$user_agent_dir/$d" ]]; then
+                cp -R "$user_agent_dir/$d" "$workdir/agent/" 2>/dev/null
+            fi
+        done
+    fi
+
+    # 覆盖 provider（如果设置了 AICODE_REGRESSION_PROVIDER_*）
+    # 用于跨供应商测试：minimax / siliconflow-anthropic / siliconflow-openai 等
+    if [[ -n "${AICODE_REGRESSION_BASE_URL:-}" ]] && [[ -n "${AICODE_REGRESSION_API_KEY:-}" ]]; then
+        # 在 models.json 里加一个 anthropic 兼容 provider
+        local api_kind="${AICODE_REGRESSION_API_KIND:-anthropic-messages}"
+        local provider_id="${AICODE_REGRESSION_PROVIDER_ID:-regression}"
+        local model_id="${AICODE_REGRESSION_MODEL_ID:-claude-sonnet-4-5}"
+        python3 -c "
+import json
+p = '$workdir/agent/models.json'
+try:
+    d = json.load(open(p))
+except: d = {'providers': {}}
+prov = d.setdefault('providers', {})
+prov['$provider_id'] = {
+    'baseUrl': '$AICODE_REGRESSION_BASE_URL',
+    'apiKey': '$AICODE_REGRESSION_API_KEY',
+    'api': '$api_kind',
+    'models': [
+        {'id': '$model_id', 'name': '$model_id', 'contextWindow': 128000, 'maxTokens': 16384}
+    ]
+}
+json.dump(d, open(p, 'w'), indent=2)
+" 2>/dev/null
+
+        # 修改 settings.json 的 defaultProvider
+        [[ -f "$workdir/agent/settings.json" ]] && python3 -c "
+import json
+p = '$workdir/agent/settings.json'
+try:
+    d = json.load(open(p))
+except: d = {}
+d['defaultProvider'] = '$provider_id'
+d['defaultModel'] = '$model_id'
+json.dump(d, open(p, 'w'), indent=2)
+" 2>/dev/null
+    fi
+
+    # 创建 stdin fifo（aicode 需要从 stdin 持续读取 RPC 命令）
+    mkfifo "$workdir/stdin.fifo" 2>/dev/null || true
+
+    # fd 11 = stdin (case 脚本可以 printf 到 &11)
+    # fd 12 = stdout+stderr 合并的 log（用于后续分析）
+    exec 11<>"$workdir/stdin.fifo"
+    exec 12>"$workdir/output.log"
+
+    # spawn 进程（默认 -ne 跳过扩展避免 conflict；要加载扩展传 false）
+    local ext_flag=""
+    if [[ "$skip_extensions" == "true" ]]; then
+        ext_flag="-ne"
+    fi
+
+    "$AICODE_BIN" $ext_flag --config "$workdir" --mode rpc --no-session \
+        <&11 >&12 2>&1 &
+    local pid=$!
+
+    echo "$workdir" > "$RESULTS_DIR/.last_workdir"
+    echo "$pid"     > "$RESULTS_DIR/.last_pid"
+
+    # 主动发一个 get_state 触发首个 response，验证 RPC 通道已就绪
+    printf '{"type":"get_state","id":"ready_probe"}\n' >&11
+
+    # 等待 response 出现（最多 5 秒）
+    local ready=0
+    for _ in $(seq 1 50); do
+        if grep -q '"id":"ready_probe"' "$workdir/output.log" 2>/dev/null; then
+            ready=1; break
+        fi
+        sleep 0.1
+    done
+
+    if [[ $ready -ne 1 ]]; then
+        kill "$pid" 2>/dev/null
+        echo "ERROR: aicode RPC 启动超时（未收到 ready_probe 响应）" >&2
+        echo "       log 前 20 行：$(head -20 "$workdir/output.log" 2>/dev/null | tr '\n' '|' | head -c 300)" >&2
+        return 1
+    fi
+
+    REGRESSION_PID="$pid"
+    REGRESSION_WORKDIR="$workdir"
+    REGRESSION_LOG="$workdir/output.log"
+    return 0
+}
+
+send() {
+    # send <json>
+    printf '%s\n' "$1" >&11
+}
+
+wait_for() {
+    # wait_for <jq-filter> [timeout_seconds]
+    local filter="$1"
+    local timeout="${2:-$AICODE_TIMEOUT}"
+    local jqbin="${JQBIN:-jq}"
+
+    if ! command -v "$jqbin" >/dev/null 2>&1; then
+        echo "ERROR: jq not found. brew install jq" >&2
+        return 127
+    fi
+
+    # 等到匹配事件出现
+    "$jqbin" -c "$filter" "$REGRESSION_LOG" 2>/dev/null | head -1
+}
+
+cleanup_rpc() {
+    # 关掉 stdin pipe + 杀进程 + 保留 log
+    [[ -n "${REGRESSION_PID:-}" ]] && kill "$REGRESSION_PID" 2>/dev/null
+    exec 11<&- 2>/dev/null
+    exec 12<&- 2>/dev/null
+
+    if [[ -n "${REGRESSION_LOG:-}" ]] && [[ -f "$REGRESSION_LOG" ]]; then
+        cp "$REGRESSION_LOG" "$RESULTS_DIR/${CURRENT_CASE}.log" 2>/dev/null
+    fi
+    [[ -n "${REGRESSION_WORKDIR:-}" ]] && rm -rf "$REGRESSION_WORKDIR" 2>/dev/null
+}
+
+trap_cleanup() {
+    cleanup_rpc
+}
+
+# ----- 断言辅助 -----
+
+assert_eq() {
+    # assert_eq <name> <expected> <actual>
+    local name="$1" expected="$2" actual="$3"
+    if [[ "$expected" == "$actual" ]]; then
+        pass "$name"
+    else
+        fail "$name" "expected: $expected | got: $actual"
+    fi
+}
+
+assert_contains() {
+    # assert_contains <name> <substring> <haystack>
+    local name="$1" needle="$2" hay="$3"
+    if [[ "$hay" == *"$needle"* ]]; then
+        pass "$name"
+    else
+        fail "$name" "expected to contain: $needle | got: ${hay:0:200}"
+    fi
+}
+
+assert_event_field() {
+    # assert_event_field <name> <jq-filter> <field-equals-value>
+    # 例：assert_event_field "model 是 anthropic" 'select(.type=="message_start")' '.message.api' "anthropic-messages"
+    local name="$1" event_filter="$2" field="$3" expected="$4"
+    local jqbin="${JQBIN:-jq}"
+    local actual
+    actual=$("$jqbin" -c "$event_filter | $field // \"\"" "$REGRESSION_LOG" 2>/dev/null | head -1)
+    if [[ "$actual" == *"\"$expected\""* ]] || [[ "$actual" == "\"$expected\"" ]]; then
+        pass "$name"
+    else
+        fail "$name" "expected $field == '$expected' | got: $actual"
+    fi
+}
+
+assert_event_count() {
+    # assert_event_count <name> <jq-filter> <min-count>
+    local name="$1" filter="$2" min="$3"
+    local jqbin="${JQBIN:-jq}"
+    local count
+    count=$("$jqbin" -c "$filter" "$REGRESSION_LOG" 2>/dev/null | wc -l | tr -d ' ')
+    if [[ $count -ge $min ]]; then
+        pass "$name ($count ≥ $min)"
+    else
+        fail "$name" "expected ≥ $min matches | got: $count"
+    fi
+}

package/scripts/regression/results/.last_pid

@@ -0,0 +1 @@
+42805

package/scripts/regression/results/.last_workdir

@@ -0,0 +1 @@
+/Users/jianjiang/.aicode-regression-42781-8293000