StackMap
Subscribe

cwc-long-running-agents vs govctl

Anthropic's harness primitives for long-running Claude agents: default-FAIL evidence gates, a fresh-context evaluator subagent and handoff hooks — each one standalone, readable file. — versus — Governance-as-code CLI for AI-assisted development: prompts and patches become RFCs, ADRs and work items with executable verification gates — reviewable, traceable, phase-gated delivery.

The curated verdict

Same core move — structural gates instead of polite prompts: govctl makes governance a CLI with verification gates; these hooks make 'done' unclaimable without opened evidence.

cwc-long-running-agentsgovctl
Stars579237
Forks6011
LanguageShellRust
LicenseApache-2.0MIT
Last activity2 months ago4 days ago
Topicscodingcoding
Curated connections32

cwc-long-running-agents — the curator's take

Read it, don't run it: an official worked example of WHY long runs fail (agents grading their own work, claiming success without evidence, losing state between sessions) with one hook per failure mode. The default-FAIL contract is the idea worth stealing. Explicitly an event demo — unmaintained, not accepting contributions — so copy the patterns into your own harness rather than depending on the repo.

govctl — the curator's take

The uncomfortable truth it addresses: with AI coding, 'done' drifts toward 'the agent stopped typing'. govctl makes governed artifacts part of the loop — RFCs state what must be true, ADRs record why, work items carry acceptance criteria, and verification guards gate completion. If your team ships AI-generated code into anything regulated or long-lived, this is the missing control plane. NOT lightweight: it's process-as-code, and process you don't enforce becomes decoration; also young (~240 stars) — expect to shape it as much as use it.