Dev shares actual orchestrator prompt for slicing work with skeptical sub-agent reviews
thatroblennon · x · 2026-11-10
thatroblennon posts the actual prompt he used minutes ago in a cybersecurity AI-harness project, showing a full orchestration method for letting agents run autonomously.
Key elements:
- /dev-delivery-ops loop: study the issue → BIG PLAN → fresh-eyes re-review → improved plan → slice it; each slice gets its own plan, re-review, execution, and an optional skeptical sub-agent review before moving on.
- Concurrency: non-overlapping slices can run in parallel; deferrable tasks must be tracked or handed to sub-agents; multiple branches merged at milestones prevent agents stepping on each other.
- Context management: keep experiments/tools/capabilities mapped and organized; write durable files (plans, tasks, docs) so sub-agents keep context; self-compact at milestones.
- /os-skillify principles: check if it's really a new skill before making one; ask one question at a time to complete the brief; preserve the source's craft and wording; write skills the way the output should read; carefully write when the skill should and shouldn't be picked — that description governs whether it gets used.
More from coding & agent
- Hermes64 plugin turns any idea into an N64-style 3D game for Nous hackathon — Teknium · 2026-11-10
- Claude rebuilds Ocarina of Time's Kokiri Forest inside Valheim, zero code written — Any-Explanation3814 · 2026-11-10
- Cortex hits 73.2% on τ³-bench banking, 18 points above best published model — multiply_matrix · 2026-11-10
- Tailscale OAuth apps go GA, letting AI agents reach private MCP servers securely — rudrank · 2026-11-10
- Medium launches TK, a writing app that plugs your notes into AI agents via MCP — LuoSays · 2026-11-10
- Developer dedicates a Mac mini as a home for his AI agents — ow · 2026-11-10