/dev/null
Sign in

Commit

80f830e

Jesse Lin@conscientiousness1/2
Herdr people, how do you run your agents? https://github.com/conscientiousness/herdr-pstack-orchestrator I use my skill to have Opus 5.5 orchestrate and DeepSeek do the work. It's really cheap. I'm on Claude Pro at $20 a month, and it's been a good fit for me over the last…
Attached image
GitHubGitHub - conscientiousness/herdr-pstack-orchestrator: Visible multi-model coding teams in Herdr: Agent Skills for Pi, Codex, Claude Code, and pstack review loops.Visible multi-model coding teams in Herdr: Agent Skills for Pi, Codex, Claude Code, and pstack review loops. - conscientiousness/herdr-pstack-orchestrator
Jesse Lin@conscientiousness2/2
My setup for the last month
ini
max_active_workers = 8
# Elapsed-time reminder only; it does not diagnose progress or cancel work.
notify_after_minutes = 30

[workers.glm]
harness = "pi"
provider = "zai"
model = "glm-5.3"
effort = "max"

[workers.ds]
harness = "pi"
provider = "commandcode"
model = "deepseek/deepseek-v4.1-flash-fast"
effort = "max"

[workers.gpt-deep]
description = "Specific correctness or timing disputes, difficult design tradeoffs, and deep review. Briefs must name the unresolved question; routine source inventory and summaries belong to other roles."
harness = "pi"
provider = "openai"
model = "gpt-6.1-sol"
effort = "xhigh"

[workers.gpt]
harness = "pi"
provider = "openai"
model = "gpt-6.1-sol"
effort = "medium"

[roles]
"feature, refactoring" = "gpt"
"bug-fix" = "gpt"
"perf-issue" = "gpt-deep"
"hillclimb" = "gpt"
"judgment and prose" = "ds"
"hardest tasks" = "gpt-deep"
"how explorer" = "ds"
"how explainer" = "ds"
"why investigators" = "ds"
"why synthesizer" = "ds"
"reflect judgment, divergent, synthesizer" = "ds"
"arena runners" = ["gpt", "ds"]
"arena cross-judge pool" = ["ds", "glm", "gpt-deep"]
"architect runners" = ["gpt-deep", "glm"]
"interrogate reviewers" = ["gpt-deep", "glm", "ds"]

commit 80f830e