← ClaudeAtlas

assumptionlisted

When an unverified project-scoped premise can change later judgment, recall, compare, and propose an ASM lifecycle result without writing the repository.
Jeis-Jw/bobbin · ★ 0 · AI & Automation · score 70
Install: claude install-skill Jeis-Jw/bobbin
# Context assumption Follow the [shared recording policy](../context/references/recording-policy.md) first. On a durable signal, resolve project settings once; only enabled owners participate automatically. User-approval instructions below describe `explicit` mode. In `auto` and `adaptive`, use policy authorization on the same validated write path; semantic attestations must remain truthful. Disabled features still allow explicit historical reads. This skill is the `context-assumption/v1` semantic owner and never writes repository bytes. - Claim only an unverified premise with project scope and a substantive basis. It must not be an observed fact or committed choice. The actual claim must equal `owner_inputs.assumption.assumption` and require `unverified_ok=true`. - Bind attestation only to `assumption_present -> /owner_inputs/assumption/assumption` and `unverified_ok -> /owner_inputs/assumption/unverified_ok` on the exact candidate. Decline OBS or DEC claims and requested-kind-only inputs. - Use metadata-first `search --signal assumption-relevant`, then selected actual-body `read`, only when a new, confirmed, refuted, or changed assumption can affect the current answer. - `confirm` requires evidence references. `refute` requires a reason, evidence references, and explicit impacted decisions. `supersede` requires both actual Assumption bodies to express the same semantic claim. `annotate` may change only meaning-preserving metadata. - Keep at most eight candidates and a 16