← ClaudeAtlas

wip-wardenlisted

Use when working on session-concurrency reporting — carries the constitutional boundary between counting sessions and watching the human, why that split protects the Reservoir Warden's trust model, the honest meaning of enforcement strict when no hook can hold a session, the flag discipline for an inexact count, and the anti-patterns
Habitat-Thinking/ai-literacy-superpowers · ★ 101 · AI & Automation · score 70
Install: claude install-skill Habitat-Thinking/ai-literacy-superpowers
# The WIP Warden Three sessions open across three repositories is not three times the work. It is one person switching, and the cost of the switch is paid in the quality of every decision made after it. > **The WIP Warden** — the concurrency gate. It counts live sessions against > the limit the human declared for themselves, and says when they are over it. **What it attacks:** thrash-switching. ## The boundary — read this before anything else > The **Reservoir Warden** watches *the human* and never gates. > The **WIP Warden** counts *sessions* and never watches the human. <!-- evidence: the Reservoir Warden's advisory-forever rule is its honesty contract. It reports proxies with honesty flags, offers one recommendation, and persists nothing — and it earns trust precisely by having no teeth. A sibling that inferred state from session counts would break that contract retroactively for everyone who had declared a chronotype. --> The split is deliberate and constitutionally significant. The moment a sibling agent starts inferring from session counts how tired someone is, or how well they are working, the Reservoir Warden's contract breaks retroactively: a human who told the harness their chronotype now has reason to wonder what else is being read from it. So the WIP Warden reports **a number, an age per session, and a comparison against a declared line**. Never a word about attention, fatigue, capacity, focus, or how anyone is doing — not as a hedge, n