Navigator plugin
Claude Code plugin for long sessions: loads documentation as needed, keeps task memory and shows how full the context window is
- 355 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The plugin's hooks and runtime intervene in every prompt: they reject some and add knowledge graph memories to the context
- The optional judge on a TypeSafe key sends the text of every prompt to a third-party API
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/navigatorDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The plugin's hooks and runtime intervene in every prompt: they reject some and add knowledge graph memories to the context
- The optional judge on a TypeSafe key sends the text of every prompt to a third-party API
This is third-party code. Review the repository files before installing.
What it does
Navigator is a Claude Code plugin built around saving context: project documentation is loaded as needed rather than all at once. Commands are plain phrases. "Initialize Navigator in this project" prepares a project, "Start my Navigator session" opens a session, and the skills in the skills folder take over, 31 in all: nav-task, nav-loop, nav-compact, nav-graph, nav-sop, nav-deep-research and more. Applied ones are included too: backend-endpoint, frontend-component, database-migration and test skills. Since version 8 the order of work is held by a runtime inside Claude Code. It rejects prompts that lack the needed preparation, limits fan-out file reads and writes every refusal to a log line. The /nav command opens a pane next to the session: context fill and trend, cost and tokens for the day, the number of rejected reads, the next step of the active task with an estimate and the knowledge graph memories for the open task. The knowledge graph stores a project's decisions, pitfalls and patterns and surfaces the relevant ones at session start. "Run until done" mode drives a task step by step with progress tracking and stagnation protection, and reply modes such as "adhd mode on" reshape replies with a phrase.
Who it is for. Claude Code users whose long sessions run into the context limit and who have to re-explain the project after every restart.
Good fit when
- A Claude Code session breaks mid-feature and the project has to be explained again
- Procedures (SOPs) and project decisions should be picked up at the start of every session
- A big task runs step by step until done and looping must be cut off automatically
- Token spend and context fill should be visible without leaving the session: the /nav command
- A topic is researched with citations: nav-deep-research builds a report with sources and writes memories to the graph
Not a fit when
- Claude Code is older than 2.1.287: the /nav pane will not open and behavior runs through the Python fallback
- You work in another agent: the plugin is built for Claude Code
- You want one small skill: Navigator brings 31 skills, hooks and a runtime
- The agent must receive prompts untouched: the Navigator runtime rejects some of them and changes how work proceeds
Example request
Run until done: add user authenticationLimitations
The plugin targets Claude Code. The runtime and the /nav pane need version 2.1.287 or newer; on older versions the same behaviors run through a Python fallback without the pane. The savings figures (from 150,000 to 12,000 tokens in the author's example) come from the author's own OpenTelemetry measurements and are not independently verified. The plugin noticeably changes how the agent works: it rejects some prompts and adds knowledge graph memories to the context.
How to disable. Disable or remove the navigator plugin through the /plugin menu in Claude Code and restart the session. Files the plugin created in the project are deleted by hand.
FAQ
How do I install Navigator in Claude Code?
Run /plugin install navigator --marketplace qf-studio/navigator on Claude Code 2.1.275 or newer and restart the program. On older versions add the marketplace first with /plugin marketplace add qf-studio/navigator.
What does /nav show?
A pane next to the session with cards: context fill, cost and tokens for the day, the fan-out read verdict and reject count, the judge's verdict on the last prompt, the next task step and knowledge graph memories. A one-line status band also sits above the prompt.
Where do my prompts go?
Nowhere beyond the machine while the optional judge is off. If you enable it through a TypeSafe key, each prompt is sent to the TypeSafe API, and the author warns about this separately.
What does Run until done mode do?
It drives a task to completion step by step: it tracks progress, finishes on a dual condition (heuristics and an explicit signal) and detects stagnation so it does not fall into an endless loop.
Are language servers required?
No. With a language server plugin such as pyright-lsp the research agent answers symbol questions through go-to-definition and references. Without one it searches with Grep as before.
Discussion
No questions or reviews yet.
Related
PluginsA dashboard for every Claude Code session on the machine: live agents, cost, history search and MCP tools for Claude.
An environment for Claude Code that keeps context across sessions via hooks, ledgers and handoffs, orchestrates agents and saves tokens
The author's personal Claude Code toolbox: meta-prompt and todo commands, skills to create skills, hooks and MCP servers, auditors and a Ralph loop
A self-improvement plugin for Claude Code, Codex and OpenCode: turns corrections into rules the agent follows in later sessions
Workflows for Claude Code: /do, /omo, /bmad-pilot and more commands, with Codex, Gemini or OpenCode writing the code through codeagent-wrapper
The Supermemory plugin for Claude Code: persistent memory across sessions with automatic prompt recall, conversation capture and a codebase index
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Navigator (this card) | no | 355 | ||
| Context Mode | no | not required | 33,535 | an MCP server that keeps tool output out of the context, while Navigator is a plugin with skills and a process runtime |
| Superpowers | no | 313,096 | a skill set for the development process: plan, TDD, review, while Navigator handles context, memory and run-until-done mode |
README in short
The page describes the problem of sessions that break from an overloaded context and compares loading with and without Navigator on the author's example. It goes on to the features of versions 5 to 8: a preference profile, the knowledge graph, run-until-done mode, reply modes and the /nav pane, then install, the optional judge and documentation links.