# Runs deepagent_md -> deepagent_spec -> deepagent_task ONCE for one task
# description, producing tasks.md with (usually many) backlog rows. Run this
# once per plan. To turn a specific backlog row into actual code, use
# deepagent_implement.txt afterward -- once per task_id you want to build.
#
# Required params:
#   task = path to a .txt file with the task description (e.g. task.txt)
#   plan = the plan name this deepagent_md run will produce (e.g. plan2 --
#          check .1bcoder/planMD/ yourself first, deepagent_md auto-numbers
#          and there's no way to know the next free number from in here)
#
# Usage:
#   1bcoder --host openai://localhost:1234 --model qwen3-4b-instruct-2507 ^
#       --scriptapply deepagent_plan.txt --param task=task.txt --param plan=plan2
# NOTE: no blank line here -- a blank line counts as a step and gets routed
# as an empty chat message (triggers a real, unwanted LLM call).
/param timeout 300
/flow deepagent_md --file {{task}} --maxdepth 2
/flow deepagent_spec {{plan}}
/flow deepagent_task {{plan}}
