gofenix/feishu-task-executor

Feishu Task Executor

You received a feishu task event. Your job: dispatch it to opencode for coding work, report results back to the task.

This is not a Nex personal task management request.

Forbidden:

• Do not use the internal task tool.
• Do not use executor_status or executor_dispatch.
• Do not inspect Nex executor runs unless the Feishu task explicitly asks for that.
• Do not answer conversationally before attempting the Feishu task execution workflow.

Step 1: Parse the task

Extract from the incoming message metadata:

• task_id: the Feishu task ID
• task_title: the task title
• task_description: the task description (may be empty)
• task_action: "created" or "rerun"

If the visible message is only /run_feishu_task <id> or /rerun_feishu_task <id>, use the ID from that command and fetch task details with lark-cli task tasks get.

Step 2: Resolve the target repo

Try in priority order:

1. Read the Feishu task custom field: Use bash to call lark-cli and extract the "仓库" custom field value:

   bash("lark-cli task tasks get --as user --format json --params '{\"task_guid\":\"<task_id>\",\"user_id_type\":\"open_id\"}'")
   

   If it returns a valid path or known alias, use it.

2. Parse @repo: from task description: Look for @repo: /path or @repo: alias-name in task_description.

3. Look up alias table: Read tasks/repos.json in the workspace. Match any word from the task title/description against the keys.

4. Ask the user: If all above fail, send a message asking "不能确定任务 <title> 对应的仓库。请回复仓库路径或别名。". When the user replies, save the alias:

   bash("echo '\"<alias>\": \"<path>\"' >> tasks/repos.json")
   

   Then re-trigger the task by changing its status from completed to in_progress.

Store the resolved repo path as REPO_PATH.

Step 3: Acquire repo lock

Use atomic mkdir to prevent concurrent execution on the same repo:

REPO_HASH=$(echo -n "<REPO_PATH>" | shasum -a 256 | cut -d' ' -f1)
LOCK_DIR="tasks/locks/$REPO_HASH"
mkdir "$LOCK_DIR" 2>/dev/null && echo '{"task_id":"<task_id>","started_at":"'"$(date -u +%Y-%m-%dT%H:%M:%SZ)"'"}' > "$LOCK_DIR/meta.json"

Execute as a single bash command:

bash("REPO_HASH=... LOCK_DIR=... mkdir ...")

If mkdir fails (lock is held):

1. Append to tasks/queues/pending.jsonl: {"task_id":"<task_id>","repo":"<REPO_PATH>","enqueued_at":"<now_iso>"}
2. Reopen the task if needed: bash("lark-cli task +reopen --as user --task-id <task_id>")
3. Add a comment: bash("lark-cli task +comment --as user --task-id <task_id> --content '排队中，前面的任务完成后自动执行。'")
4. STOP — do not proceed further.

Step 4: Execute opencode

Reopen the task if it is completed (unless already done in Step 3):

bash("lark-cli task +reopen --as user --task-id <task_id>")

Run opencode with the task title and description as the prompt:

bash("cd <REPO_PATH> && opencode run --dangerously-skip-permissions '<task_title>: <task_description>' 2>&1", timeout: 1800)

Capture both exit code and output:

if cd "<REPO_PATH>" && opencode run --dangerously-skip-permissions "<task_title>: <task_description>" 2>&1; then EXIT_CODE=$?; else EXIT_CODE=$?; fi
echo "EXIT_CODE=$EXIT_CODE"

Use the larger timeout (at least 1800) since coding tasks can take a while. Check the EXIT_CODE from the output — 0 means success.

Step 5: Report results

If opencode succeeded (exit code 0):

1. Update task status to completed: bash("lark-cli task +complete --as user --task-id <task_id>")
2. Add a comment with output summary (first 500 characters, trim to avoid API limits):

   bash("lark-cli task +comment --as user --task-id <task_id> --content '执行完成。\n\n<output_summary>'")
   

3. Notify the user with a brief message + task link.

If opencode failed (non-zero exit code):

1. Keep task status as in_progress (so user can see and re-open it later).
2. Add a comment with the error output: bash("lark-cli task +comment --as user --task-id <task_id> --content '执行失败。\n\n<error_output>'")
3. Notify the user with the error info.

Step 6: Release lock + dequeue

Release the lock:

bash("rm -rf tasks/locks/$REPO_HASH")

Check if there are queued tasks for the same repo:

bash("cat tasks/queues/pending.jsonl 2>/dev/null")

If entries exist for this repo:

1. Dequeue the first matching entry
2. Extract its task_id and other fields
3. Format a new task content as /run_feishu_task <dequeued_task_id>
4. Set _from_feishu_task: true in metadata (through the task_action mechanism)
5. Return to Step 2 for this dequeued task

If no queued tasks, you are done.