xiaojia-skills（小甲技能）

Xiaojia Skills

Use the bundled scripts to inspect available projects and skills, then call the deployed JustAI openapi endpoint in two steps:

1. list_projects.py lists your available资料库 folders/projects
2. list_skills.py lists your available skill IDs
3. chat.py submits the task and returns conversation_id
4. chat_result.py polls chat_result until the task is completed or failed

This keeps the interface stable for slower branches such as card generation, plan generation, notes generation, and image generation.

When running inside Claude Code, ${CLAUDE_SKILL_DIR} resolves to this skill directory. Use that path when invoking the bundled scripts. In Codex or a plain shell, run the same scripts from the installed skill directory.

Workflow

1. Ensure these environment variables are set before using the script:

- JUSTAI_OPENAPI_BASE_URL - JUSTAI_OPENAPI_API_KEY - Optional: JUSTAI_OPENAPI_TIMEOUT

1. If the task should be scoped to a specific资料库, run scripts/list_projects.py first and choose one or more project_id.
2. If the task should preload a specific skill, run scripts/list_skills.py first and choose one or more skill_id.
3. Run scripts/chat.py with --message, optional --conversation-id, optional repeated --project-id, and optional repeated --skill-id.
4. Preserve the returned conversation_id.
5. Run scripts/chat_result.py --conversation-id ... to poll the task result.
6. Read the returned JSON and use it directly:

- branch tells which real path ran, such as collect_info, confirm_info, generate_plan, generate_notes, or generate_image - result is the primary payload - text is the human-readable summary or fallback text - conversation_id must be preserved for follow-up turns

1. If the user asks to continue an existing conversation, pass the previous conversation_id back into scripts/chat.py together with the new --message.
2. For confirm_info results:

- To accept the card and continue, send a natural-language follow-up such as 这些信息没问题，继续生成方案. - To revise the card, send a natural-language correction such as 预算改成3万，目标用户改成25到30岁女性，请更新资料卡片. - Do not attempt structured form submission yet; the openapi layer currently supports message-based continuation only.

Commands

List projects:

python3 "${CLAUDE_SKILL_DIR}/scripts/list_projects.py"

List skills:

python3 "${CLAUDE_SKILL_DIR}/scripts/list_skills.py"

Run a new turn:

python3 "${CLAUDE_SKILL_DIR}/scripts/chat.py" --message "帮我做一份小红书运营方案"

Poll the result:

python3 "${CLAUDE_SKILL_DIR}/scripts/chat_result.py" \
  --conversation-id "existing-conversation-id"

Continue an existing turn:

python3 "${CLAUDE_SKILL_DIR}/scripts/chat.py" \
  --conversation-id "existing-conversation-id" \
  --message "继续展开第二部分"

Continue after a confirmation card:

python3 "${CLAUDE_SKILL_DIR}/scripts/chat.py" \
  --conversation-id "existing-conversation-id" \
  --message "这些信息没问题，继续生成方案"

Revise a confirmation card in natural language:

python3 "${CLAUDE_SKILL_DIR}/scripts/chat.py" \
  --conversation-id "existing-conversation-id" \
  --message "预算改成3万，目标用户改成25到30岁女性，其他不变，请更新资料卡片"

Run a turn scoped to a selected project/folder:

python3 "${CLAUDE_SKILL_DIR}/scripts/chat.py" \
  --project-id "fld_demo" \
  --message "请参考这个资料库里的内容，帮我写一篇港理工校园生活图文笔记"

Run a turn with a manually selected skill:

python3 "${CLAUDE_SKILL_DIR}/scripts/chat.py" \
  --skill-id "skill_demo" \
  --message "使用这个技能继续分析"

Run a turn with both selected project and skill:

python3 "${CLAUDE_SKILL_DIR}/scripts/chat.py" \
  --project-id "fld_demo" \
  --skill-id "skill_demo" \
  --message "优先参考这个资料库并使用这个 skill 帮我生成方案"

Override the timeout for slower branches:

python3 "${CLAUDE_SKILL_DIR}/scripts/chat.py" \
  --message "我是做敏感肌护肤的品牌，先帮我整理资料卡片"

Control polling interval:

python3 "${CLAUDE_SKILL_DIR}/scripts/chat_result.py" \
  --conversation-id "existing-conversation-id" \
  --poll-interval 2 \
  --timeout 300

Guardrails

• Prefer the scripts over hand-written curl so submit and polling stay consistent.
• When the task depends on a specific资料库 or手动技能, list them first and pass the exact IDs. The openapi request now统一使用 project_id 和 skill_id 两个字段，值都是字符串数组。
• Keep conversation_id from the last successful submit response and reuse it for polling and follow-up requests.
• Treat result as the primary machine-readable payload. Use text only as fallback when result is plain text.
• confirm_info currently continues through natural-language follow-up on the same conversation_id; there is no structured form_data submission path in openapi yet.
• If the endpoint returns status=failed, surface message directly instead of retrying blindly.