DAGExecutor¶

Role: Owned execution backend — DAG-based workflow executor GitHub: ProtocolWarden/DAGExecutor License: AGPL-3.0-or-later

What It Does¶

DAGExecutor drives multi-node AI task graphs with topological scheduling and concurrent layer execution.

Node types:

Node	What it does
`agent`	Claude Code or Codex CLI subprocess — receives `goal_text` verbatim
`bash`	Shell command (no shell features — list-form exec via `safe_run`)
`script`	Python script written to a temp file and executed
`loop`	Static poll (repeat until stdout contains `DONE`) or dynamic fan-out (one run per item, concurrent)
`gate`	Human file-based approval — advances (`approved`) or halts (`rejected`)

Graph engine: rustworkx — Rust-backed DAG with cycle detection, topological sort, and layer-based concurrent execution via ThreadPoolExecutor.

worker_backend selects the CLI used for agent nodes:

claude_code (default) — claude --message … --output-format json --no-auto-commits
codex_cli — codex --model … --approval-mode full-auto -q …

Agent node commands are built as list[str] directly — no shell quoting or string construction.

All subprocess calls use core_runner.process.safe_run():

start_new_session=True — child owns its own process group
os.killpg(SIGKILL) on timeout — kills full descendant tree
No shell=True — commands are list-form; string commands split via shlex.split()

D1: goal_text reaches agent nodes verbatim (node command appended as --append-system-prompt)
Cycle detection fires before execution; malformed graphs are rejected at load time
Gate nodes can only advance (approved) or halt (rejected) — no silent pass-through

Replaces the Archon external integration (retired 2026-05-18, ADR 0005).