agenctx CLI
agenctx keeps project rules and warnings in the repo, and agents find the relevant ones by search instead of reading everything.
- 291 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The tool works locally and writes files to `.agenctx/` and agent guides.
- Entries proposed by an agent are untrusted until human approval.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/agenctxDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The tool works locally and writes files to `.agenctx/` and agent guides.
- Entries proposed by an agent are untrusted until human approval.
This is third-party code. Review the repository files before installing.
What it does
agenctx is a local context management system for agents, versioned with the repository. A human adds rules, decisions, warnings, test requirements and security constraints to `.agenctx/` in channels such as `design`, `security`, `testing` and `warnings`. `agenctx dump` generates short guides for Codex, Claude and Cursor that teach the agent to search, view previews and open the needed entries. An agent may propose at most two entries per session and they become trusted after a human decision. Each session leaves a hash-chained receipt of what was served and in what order. The `agenctx ui` interface runs on 127.0.0.1.
Who it is for. Teams whose `AGENTS.md` has grown into a dump of rules and now fills the agent context with irrelevant text.
Good fit when
- The project rules file got long and an important warning is buried.
- A security task should receive only security rules.
- An agent found a useful pattern and it must be proposed for review, not written into trusted rules.
- You need to audit exactly which instructions the agent received in a session.
Not a fit when
- There are few rules and one `AGENTS.md` is enough.
- You need a hosted company-wide knowledge base.
- You need a standalone coding agent, which this tool is not.
Example request
Add the warning "never edit generated API files", then search the context for SQLite entries and show the previews.Limitations
Node.js 18 or newer is required. Knowledge lives in the `.agenctx/` directory, which must be committed and maintained by people: without an owner entries go stale. Agents find context only through generated guides and the CLI, so the client must read them. Agent proposals stay untrusted until a human decides. The last repository commit is from October 2026. The local interface is protected by a launch token but starts on your computer and is not meant for shared access.
How to disable. Delete the `.agenctx/` directory and the generated guides in agent files. Remove the package with `npm uninstall --global agenctx`.
FAQ
How is this better than a single `AGENTS.md`?
The agent receives only the relevant channel instead of the whole file. Entries have a lifecycle and old ones go to the archive.
How do I add an entry?
With `agenctx add`, for example `agenctx add warn "Never edit the generated API files"`. Types include `warn` and `decision`.
Can an agent change trusted rules?
No. It proposes up to two entries per session and they stay untrusted until a human approves them.
Is there telemetry?
No. The tool is local, dependency-free and has no telemetry.
What is a session receipt?
A hash-chained record of which context was served to the agent and in what order. It lets you audit a session.
Discussion
No questions or reviews yet.
Related
CLI toolsAgentic Substrate
Agentic Substrate (claude-user-memory)
A short reversible rule set for an agent: research the unclear, plan proportionally, verify the outcome and keep checked knowledge
One config for Claude Code and Codex in every repo: shared rules, skill packs, second-model review and a destructive-command guard
Agents Can Communicate (ACC)
Agents Can Communicate
ACC passes questions, reviews and handoffs between Claude Code, Codex and other client sessions on one machine
A skill for messaging between Claude Code, Codex, Gemini CLI and other CLI agents through a shared local SQLite database, written in bash
A file-based context system keeps tasks, decisions and learnings between sessions so an agent never starts from zero.
Agav is a terminal coding agent that reads a repository, edits files, runs tests and switches between seven model providers.
README in short
The README explains the problem with static instruction files, the library model, six working steps, the local UI and npm installation. It includes session examples.