Launching today
Stagent
Drive Claude Code through long tasks it would otherwise drop
20 followers
Drive Claude Code through long tasks it would otherwise drop
20 followers
Claude Code is great at starting long tasks — bad at finishing. It self-approves, patches symptoms, fakes TDD, stops at "code written." Stagent drives Claude Code through any state machine you define (e.g. plan → verify → review → ship). Different agents per stage - it can't self-approve or bail halfway. Describe your own workflow in plain English with /stagent:create, or fork one from the cookbook: stagent.worldstatelabs.com/cookbook Plus: live viewer, cross-machine resume.
Stagent
I like the idea of separating responsibilities across different agents instead of relying on a single agent for the entire workflow . One thing I’m curious about — if the agents share similar context and reasoning patterns, how do you prevent them from effectively behaving like the same agent in different stages?
Stagent
@anushkaamittal Great question — it’s the thing we worried about most when designing stagent.
The differentiation comes from three concrete things:
1. Different instruction files per stage. Planning reads planning.md, executing reads executing.md, QA reads qa-ing.md. Same underlying agent shell, but the active prompt is the stage’s canonical doc — so the reasoning frame is genuinely different each turn.
2. Different inputs and artifacts. The planner only sees the topic + workflow shape. The executor only sees the plan. The reviewer only sees the diff + the executor’s report. Each stage is context-starved on purpose, which forces it to play its role instead of re-deriving everything.
3. Different exit contracts. Each stage has to emit a report with required frontmatter (status, next, etc.) before the state machine will let it transition. A planner that tries to “just implement it” can’t — the schema rejects the artifact.
So yes, the model may be the same, but the agent — defined as (prompt + context window + allowed outputs) — is structurally different. The state machine is what keeps them honest; without it, you’re right that they’d collapse into one chatty generalist.