← ClaudeAtlas

gabe-walklisted

Human-eye verification — BRIEF the walker (why · what changed · itinerary · verdict meanings), then record who·when·result·evidence to walks.jsonl. Records, never judges.
khujta/gabe-suite · ★ 0 · Data & Documents · score 63
Install: claude install-skill khujta/gabe-suite
# Gabe Walk — the witness **Usage:** `/gabe-walk <feature|procedure> [pass|fail|partial]` ## Gabe execution contract (E1–E7) This skill runs under the suite execution contract — E1 EVIDENCE · E2 RUN-BEFORE-✅ · E3 NO SILENT DOWNGRADE · E4 REUSE FIRST · E5 STATE SYNC · E6 MISSING ANCHOR = STOP · E7 REPORT WHERE — floors, not ceilings; a skill's own gate may be stricter, never looser. Full text: `../gabe-docs/references/execution-contract.md` (if that file is missing, E6 applies — STOP). ## The intention *Nothing in git, junit, or coverage knows whether a person opened the app and looked. This skill records that one fact and nothing else. It records — it never judges: a half-walk is a `partial`, a bad walk is a `fail`, and a subject with zero records renders NEVER-walked (red) until someone actually walks it. A synthesized witness is worse than an honest gap.* *But a witness who was never told what they are witnessing produces a worthless record. So the walk has two halves: BRIEF first, record second. The briefing translates what is already on the record into "here is what you are looking at, and why your eyes are needed" — derived from committed data, never invented. Judging stays out of both halves.* ## Procedure 1. **Subject** — the feature slug or procedure name being walked (e.g. `transaction`, `deploy-rollback`). If the argument is missing, ask. For an entity walk the subject is the BARE adoption.json slug; `/gabe-adopt` approvals record `adopt:<slug>` — the center