How to use the Ask Navigator agent skill for Claude

What is an agent skill, and what does ask-navigator do?

An agent skill is a folder containing a SKILL.md file that gives an AI agent specific instructions for a task. When the skill is installed, the agent can follow those instructions when a request matches.

Ask Navigator is for efforts wrapped in fog: the way from here to the destination is not yet visible. The skill checks two things. Can the destination be stated in one sentence? Can the first three decisions be stated precisely, even if none can be answered yet? If either answer is no, the effort is foggy and the skill charts it.

The skill produces decisions, not deliverables. It does not write the feature. When the path is clear, it hands the settled decisions to the separate /oh-my-claudecode:launch skill as a mission brief.

Install ask-navigator from a ZIP file

Download the ZIP for the ask-navigator skill from this page. Then install it using one of the two locations below, or upload it in the Claude apps.

Option 1 is a personal install that is available across your projects: unzip the download into ~/.claude/skills/. Option 2 is a project install: unzip it into the .claude/skills/ folder of a specific project.

Option 3 is to upload the ZIP in the Claude apps under Settings, then Capabilities, then Skills.

• Personal: ~/.claude/skills/
• Project: .claude/skills/ inside your project
• Claude apps: Settings → Capabilities → Skills

Trigger ask-navigator with natural requests

The skill's argument hint lists four ways to invoke it: a loose idea, residual questions handed over by launch, a map link or number, or no argument to continue the open map.

The skill's metadata sets disable-model-invocation to true, so you should invoke it deliberately rather than expect it to start on its own. Use the slash command form when you want it to run.

• Loose idea: give a vague feature or spec in one line.
• Residual questions: pass the open questions that launch's fog gate handed over.
• Map link or number: continue a specific map.
• No argument: pick up the open map and work the next frontier ticket.

How a session works: charting and working the map

When you start from a loose idea, the skill first runs a drydock audit in report-only mode. Findings do not block charting. They are recorded in the map's Notes so they can be handled later when the effort enters delivery.

Next, the skill uses deep-interview to name the destination. You must confirm that destination explicitly. Then it maps the frontier with breadth-first rounds, asks the open decisions, and presents a proposed chart with tickets and blocking edges. You sign the chart before tickets are created.

The map lives in a single issue labelled navigator:map on your repo's issue tracker. Each decision is a child issue. The skill stops at the end of a charting session and tells you the map link, how many tickets are on the frontier, and to run the skill again for the next decision.

• The agent drafts the facts; you make the decisions.
• The agent should not answer questions that belong to you.
• Ruled-out work stays out of the map unless you reopen it.

What ask-navigator needs before it runs

The skill needs a repo issue tracker that the agent can reach. The source says the tracker is the only map medium, with no local fallback. If the tracker is unreachable, the skill stops and reports a blocker.

It calls other oh-my-claudecode skills: deep-interview, launch, and drydock. The file also references a scripts/shipyard-audit.mjs script, run with node. That script must exist in your repo for the audit step.

The skill runs scripts and uses the network.

• A reachable repo issue tracker
• The oh-my-claudecode skills deep-interview, launch and drydock
• scripts/shipyard-audit.mjs in the repo, run via node, for the audit step

Safety notes and when not to use it

Because the skill runs scripts and uses the network, review the files before you install it, especially any scripts it calls in your repo.

Do not use it for clear, well-defined tasks. The skill itself says no map is needed in that case and sends you to launch. It is also a poor fit for fully autonomous decision-making, since the skill treats an agent answering its own questions as a broken process.

• Clear tasks: use /oh-my-claudecode:launch instead.
• No issue tracker: the skill cannot run.
• Fully autonomous work: the skill expects you to answer the questions.