Orchestrate AI worker tasks in a DAG with budget caps and tmux panes.
npx skills add https://github.com/quickcall-dev/skills --skill dag-fleet
Please help me install this Agent Skill. Skill: dag-fleet Source: https://github.com/quickcall-dev/skills/tree/main/skills/dag-fleet Command: npx skills add https://github.com/quickcall-dev/skills --skill dag-fleet
This Skill eliminates the complexity of coordinating multiple long-running AI worker tasks that must run in dependency order while staying within per-worker and fleet-wide budgets.
depends_on workers complete.max_budget_usd) and optional fleet total cap (max_budget_fleet) to control spending.claude -p), Codex (codex exec), and Pi (pi -p) with provider-specific settings and tool restrictions.Use the dag-fleet skill to launch a fleet by creating a fleet.json with your workers and dependencies, then run: npx skills add quickcall-dev/skills --skill dag-fleet
High-intent search queries and answers about installing and using this skill.
Orchestrating parallel AI agents with dependency ordering in tmux is done by defining a DAG in fleet.json. The system applies a topological sort to launch workers in layers, ensuring each task starts only after its depends_on workers complete successfully.
Yes, you can enforce per-worker budget caps for Claude and Codex CLI tasks. The system applies hard max_budget_usd limits per worker and an optional max_budget_fleet cap for the total fleet, preventing runaway spending across multi-provider execution.
Managing persistent tmux panes for long-running AI worker tasks is handled by running each worker in its own tmux window. This provides persistent visibility and manual control, allowing operators to attach, monitor, and inspect ongoing sessions across disconnects.
Yes, it supports multi-provider workflows with Claude, Codex, and Pi. The system executes tasks using provider-specific commands like claude -p, codex exec, and pi -p, applying individual provider settings and tool restrictions per worker.
To relaunch a failed worker in a DAG without restarting the entire fleet, use the operator-managed lifecycle scripts. The relaunch-worker script targets individual failed tasks while preserving the state of completed dependencies in the tmux session.
Specifying absolute output paths in prompt files for AI worker tasks is required for reliable dependency resolution. Downstream workers in the DAG topo-sort layer read these absolute paths to locate upstream results, preventing broken dependencies during parallel execution.
claughinghouseCreates secure, isolated sandboxed code execution environments for AI workflows.
shanevRun long-running tasks in detached tmux sessions with persistent logs.
EdWrldCoordinate CLI workflows with done-signal files and optional locks.
t3taExecute LLM-generated DAG plans with dependency resolution and state persistence.