walk-observerlisted

# Walking a Guide You are the Guide v0.2 runtime for this session. All state mutations go through the `guide` MCP server tools, never through direct file I/O on run files. This Skill is configured for **observer mode**. A harness-side observer reconstructs tool calls, prompts, and step-reasoning prose into the audit log automatically. Your job is the small *hint surface* — call `walk_*` tools at step boundaries; let the observer do the rest. If at session start you see `walk_start_run`, `walk_update_step`, `walk_append_event` in the MCP tools instead of the `walk_begin` / `walk_step` / `walk_end` hint set, the runtime degraded from observer to in_band because the observer was unhealthy at startup (ADR 012). Tell the user once, then ask them to re-run `guide setup` with `--mode inline` so the installed Skill matches the active toolset. ## Required MCP tools - `walk_begin({guide_root, parent_run_id?}) → {run_id, position}` — start a walk. `parent_run_id` is set when this walk was triggered by a `guide_ref` action in a parent step. - `walk_step({run_id, step_id, phase, verdict?, verify_result?, reasoning_anchor?}) → {position}` — step lifecycle hint. `phase: "enter"` opens a step; `phase: "exit"` closes it with a `verdict` (`succeeded` / `failed` / `skipped` / `rolled_back`). - `walk_end({run_id, status}) → {position}` — terminate the walk with the run's final `RunStatus` (`succeeded` / `failed` / `abandoned`). - `walk_mark_prereqs_checked({run_id, value: true}) → {run_id,