@nseng-ai/objectives

ns Objectives, packaged as an installable ns extension.

An Objective is a durable, checked-in planning record for work that outlives a single agent session — multi-session, multi-branch, or multi-PR work. Your coding agent reads and maintains the record; you review it in ordinary diffs. Objectives live as Markdown under .ns/objectives/ in your repository.

This package is an ns extension: it is installed on top of the bare-core @nseng-ai/ns CLI, not bundled with it.

Install

Requires @nseng-ai/ns installed and a repository already activated with ns init:

npm install -g @nseng-ai/ns
ns init --harness claude-code                  # once per repository
ns extension install npm:@nseng-ai/objectives

A bare-core ns install does not include ns objective. Installing this extension:

1. Records @nseng-ai/objectives in ns.toml and activates it for your configured harness.
2. Adds the ns objective command surface.
3. Adds an Objectives instruction block to AGENTS.md, teaching agents to check ns objective list before non-trivial work and to use the Objective skills.
4. Provisions the ten Objective skills into your harness's skill root — .claude/skills/ for Claude Code: objective, objective-autorun, objective-close, objective-create, objective-critique, objective-next, objective-refresh, objective-retro, objective-runner-step, objective-update.

ns extension install writes files but never commits — review and commit them yourself. After the first install, ns extension update refreshes the provisioned artifacts when the extension changes.

Lifecycle

Drive an Objective through create → next → update → close by asking your agent in natural language; each step maps to a skill:

| Step | Ask your agent | Skill | | ----------- | -------------------------------------------------------------------- | ------------------ | | Create | "Create an objective for migrating our API layer to typed handlers." | objective-create | | Advance | "What's next on the api-typed-handlers objective?" | objective-next | | Update | "Update the objective with what we landed." | objective-update | | Close | "Close the objective." | objective-close |

Because the record is checked in, a fresh agent session starts with the full history of intent instead of an empty context window. Progress is evidence-linked prose, not a checkbox dashboard; closing records the rationale and marks the record closed without deleting anything.

CLI

The extension adds a deterministic, read-only ns objective surface for listing and inspecting records:

ns objective list           # open Objectives (auto-detects trunk; reads .ns/objectives/)
ns objective show <slug>    # inspect one record

ns objective list is zero-config; closed records drop out of the default listing.