← ClaudeAtlas

walklisted

Walk a Guide. Activate when the user asks to walk, run, execute, perform, or resume a Guide; or when they reference a multi-step workflow by name and want it driven end-to-end. Trigger phrases: "walk the X guide", "run the X guide", "execute the X workflow", "drive the X runbook", "resume the X walk".
agentguides/claude-plugin · ★ 0 · AI & Automation · score 74
Install: claude install-skill agentguides/claude-plugin
# `walk` — guide walk Skill router This Skill routes to the mode-specific walk Skill based on the runtime environment. The guide walk runtime supports two audit-mode surfaces — **observer** and **in_band** (a.k.a. **inline**) — installed side-by-side as `walk-observer` and `walk-inline`. Which one to use is a per-walk choice driven by the MCP server's current mode. ## Resolve the active mode Run one short shell check to determine which surface the MCP server is serving for this walk: ``` echo "GUIDE_AUDIT_MODE=${GUIDE_AUDIT_MODE:-<unset>}" ``` Then route: 1. **`observer`** — activate the `walk-observer` Skill and follow its protocol (the hint surface: `walk_begin` / `walk_step` / `walk_end`). 2. **`in_band`** — activate the `walk-inline` Skill and follow its protocol (the state-direct surface: `walk_start_run` / `walk_append_event` / `walk_update_step` / `walk_set_status`). 3. **`<unset>`** (or anything else) — STOP and report `audit mode not configured; cannot walk`. The harness should have set `GUIDE_AUDIT_MODE` via the per-walk env overlay; an unset value means the harness setup was bypassed. This router itself does not call any MCP tools — its only job is to hand off to the right concrete Skill. After activation, all walk mechanics live in `walk-observer` / `walk-inline`. ## Why two Skills? The two surfaces differ in **who produces audit events**: - **observer**: the harness observes the agent's tool-call stream and reconstructs audit events.