# task sections > **Prerequisites:** Please read `../../lark-shared/SKILL.md` to understand authentication, global parameters, and security rules. ## Scope and Operation Guide This reference organizes Task `sections` workflows by Catalog method. It currently covers locating sections and listing their tasks. See the [main Skill](../SKILL.md) for group terminology and ownership routing. The current CLI Catalog/schema is authoritative for parameters, permissions, and response fields. | User goal | Read | |----------|----------| | List sections in a resource or locate a section by name | [sections.list](#sectionslist) | | List tasks in a selected section | [sections.tasks](#sectionstasks) | ## Shared Rules 1. Follow the main Skill's discovery flow: run `lark-cli task --help` and `lark-cli task sections --help`, confirm the method, and read its schema before executing a native command. 2. Use `--as user` when querying the current user's personal "My Tasks". For other operations, confirm identity and permissions through the shared rules and the method's help/schema. 3. For queries that support pagination, use `--page-all --page-limit 0` when a complete listing is required. This avoids the default page limit. If the query is interrupted or partially fails, report that results are incomplete; do not conclude that a section or task does not exist. ## sections.list List sections in "My Tasks" or a specific tasklist, or resolve a section name to its GUID. ### Recommended Commands **List sections in "My Tasks"** ```bash lark-cli schema task.sections.list lark-cli task sections list --resource-type my_tasks --as user --page-all --page-limit 0 ``` **List sections in a specific tasklist** ```bash lark-cli schema task.sections.list lark-cli task sections list --resource-type tasklist --resource-id "" --as user --page-all --page-limit 0 ``` ### Parameters Read `lark-cli schema task.sections.list` for the full parameter structure. Choose the owning resource as follows: - "My Tasks": use `my_tasks`; no tasklist GUID or `resource_id` is needed. - A specific tasklist: use `tasklist` and pass its GUID through `--resource-id`. ### Workflow 1. Determine ownership from context; ask for the owning resource if it remains unclear. Use a known tasklist GUID directly. If only the tasklist name is available, follow the main Skill's tasklist lookup flow first. Do not use the section name as a tasklist search term. 2. Execute the command for the owning resource. 3. Match `name` in the returned `items`, and use the selected item's `guid` as `section_guid`. If names are duplicated, show candidate GUIDs and their known ownership so the user can select the target. 4. Report the sections or pass the selected GUID to `sections.tasks` when the user requests their tasks. If a complete listing has no match, report that the section was not found in that resource; do not automatically scan other tasklists. ## sections.tasks List tasks in a selected section. ### Recommended Commands ```bash lark-cli schema task.sections.tasks lark-cli task sections tasks --section-guid "" --as user --page-all --page-limit 0 ``` ### Parameters Read `lark-cli schema task.sections.tasks` for the full parameter structure. Pass the target section's GUID through `--section-guid`. ### Workflow 1. Use a known section GUID directly. If only the section name is available, locate it with [sections.list](#sectionslist) first. A known section GUID does not require another section or tasklist lookup. 2. Execute the command to list the section's tasks. 3. Present the returned tasks as requested. Report a complete result only after pagination finishes; otherwise, state that the results are incomplete. ## Adding Operation Guidance For each added operation, create a `## sections.` chapter using its Catalog method identifier and update the operation guide. Start with a short description, then use the Task reference convention: **Recommended Commands → Parameters → Workflow**. Keep shared rules at the start of this reference and operation-specific differences in the relevant chapter. Commands must show schema discovery before execution. Use Parameters for operation-specific choices and a schema pointer; leave complete parameter, permission, and response definitions to the schema.