v1.2.9

v1.2.9

• 严格校验 OpenAI-compatible 模型最终 content，HTTP 200 但 content 为空不再算健康通过。
• 仅返回 reasoning_content 时提示关闭 thinking 或增大 max_tokens，不再作为有效多模型讨论输出。
• 复用统一 payload 构造和响应解析逻辑，工作流调用不再静默接受空输出。
• 更新 DeepSeek / GLM 默认接入 preset、模板、README 和参考文档。
• 补充 reasoning-only mock 测试、完整用户流程模拟和敏感信息审计。

v1.2.8

v1.2.8

重点修正

• 新增一键接入命令：setup_model_pool.py connect 。
• 新增 --store auto：macOS 使用 Keychain，Windows 使用 DPAPI 用户级加密，Linux 使用 Secret Service；不可用时提示环境变量兜底，不默认明文保存。
• 健康检查和工作流模型调用支持 secret_ref，不再只依赖环境变量。
• 首次安装引导改为“帮我接入模型 + 安全保存 Key”的用户路径。
• README、SKILL、模板和参考文档同步更新跨平台安全 Key 存储说明。
• 新增安全 Key 接入模拟测试，验证模型池 JSON 不写真实 Key。

验证

• python3 -m py_compile scripts/*.py
• python3 scripts/setup_model_pool.py connect deepseek --store auto --dry-run
• python3 scripts/setup_model_pool.py connect deepseek --store env --api-key-env BLCAPTAIN_TEST_DEEPSEEK_KEY --config tests/runs/secure-connect-model-pool.json
• python3 scripts/check_model_pool.py --config tests/runs/secure-connect-model-pool.json
• python3 scripts/quick_validate.py
• python3 scripts/simulate_user_flow.py
• Codex 宽泛痛点 Cut-to-Go 工作流生成并通过 PRD 校验
• 敏感信息、个人路径和无关参考痕迹审计通过

v1.2.7

v1.2.7

重点修正

• 新增首次运行模型 Agent 引导：安装后先展示支持的模型类型、推荐本地配置路径、密钥安全规则和健康检查命令。
• 新增 scripts/setup_model_pool.py，支持 --doctor 和 --init，帮助新用户完成模型池 Bootstrap。
• 新增模型 Agent 清单和首次运行参考文档，明确 OpenAI-compatible、CLI Agent、本地模型、Codex 主持的边界。
• 新增痛点簇分析与 Cut-to-Go：宽泛 Watch/Pivot 机会会先收窄为可验证的最小商业切口，再决定是否生成 PRD。
• 新增 Codex 宽泛痛点演练样本，验证可收窄到“新用户安装与多模型配置护航”并生成工程实施 PRD。

验证

• python3 -m py_compile scripts/*.py
• python3 scripts/quick_validate.py
• python3 scripts/simulate_user_flow.py
• python3 scripts/setup_model_pool.py --doctor --config tests/runs/new-user-model-pool.json
• Codex 宽泛痛点 Cut-to-Go 工作流生成并通过 PRD 校验
• 敏感信息、个人路径和无关参考痕迹审计通过

v1.2.6

v1.2.6

Added

• Added model-pool Bootstrap as the hard first stage for the end-to-end workflow.
• Added CLI and OpenAI-compatible connection candidate discovery in model health checks.
• Added structured --json-output support for model health checks.
• Added Pivot-to-Go loop for opportunities that need a narrower cut before PRD generation.

Changed

• Stopped run_opportunity_workflow.py at ConfigRequired when no external model passes the skill-owned model pool health check.
• Restricted discussion execution to models registered in the model pool and passing health checks.
• Updated Chinese and English docs to clarify that discovered candidates do not count as usable models until configured and checked.
• Replaced plaintext-looking API key export examples with safer environment-variable guidance.

Verified

• python3 -m py_compile scripts/*.py
• python3 scripts/quick_validate.py
• python3 scripts/simulate_user_flow.py
• ConfigRequired workflow stops before evidence scan, discussion, Gate, or PRD generation
• Go workflow generates and validates commercial + engineering PRD
• Pivot-to-Go workflow preserves reverse evidence, reruns Gate, and validates generated PRD
• Sensitive information and reference-trace audit

v1.2.5

Added

• Added a simpler Codex-assisted model setup path for new users.
• Documented friendly prompts for connecting DeepSeek, GLM, Claude CLI, Gemini, Grok, and local CLI models.

Changed

• Made Codex-assisted setup the recommended path in Chinese and English README files.
• Updated model configuration guidance, health-check output, and simulation fixtures so users do not need to start by editing JSON manually.
• Kept manual JSON configuration as the advanced fallback for users who want full control.

Verified

• quick_validate passed.
• simulate_user_flow passed.
• Python compile checks passed.
• New-user install simulation passed.
• Sensitive information and reference-trace audit passed.