Skillget

agentic-workspace-core

agent
Security Audit
Fail
Health Warn
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Fail
  • spawnSync — Synchronous process spawning in bin/agentic-workspace-core.mjs
  • fs.rmSync — Destructive file system operation in bin/agentic-workspace-core.mjs
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

Repository-native agent memory, skills, AGENTS.md, llms.txt, and knowledge workflows for coding agents.

README.md

Agentic Workspace Core

npm version
CI

Repository-native memory and workflow infrastructure for coding agents.

Agentic Workspace Core installs a shared AGENTS.md, durable project knowledge in docs/, reusable agent skills, generated llms.txt navigation, handoff conventions, and validation scripts. It keeps agent context inspectable, versioned, and reviewable instead of scattering project memory across hidden chats, stale prompts, and tool-specific rule files.

It is not an agent runtime, framework, vector database, or Obsidian vault. It is a portable core layer you install into a repository so Codex, Claude Code, and other coding agents can work from the same project-native source of truth.

Why Use It

Use Agentic Workspace Core when you want agents to:

  • start from one clear repository instruction entrypoint: AGENTS.md
  • remember project facts through reviewed Markdown docs, not chat history
  • use focused skills for software development, frontend UI, research, handoffs, skill authoring, and project knowledge
  • rebuild generated navigation instead of trusting stale generated context
  • archive old agent rules and memory files into legacy/ before replacing them
  • keep every install, update, and memory change visible in git

Quick Start

Run from the root of the repository where agents will work:

npx agentic-workspace-core@latest init --dry-run
npx agentic-workspace-core@latest init
npm run knowledge:check

init --dry-run prints the exact files that would be created, replaced, or archived. Use it first on existing repositories.

Do not clone this GitHub repository into the target project. Install through the npm CLI so the package places its managed files at the project root.

If the working directory is uncertain, resolve the repository root explicitly:

repo_root="$(git rev-parse --show-toplevel)"
npx agentic-workspace-core@latest init --target "$repo_root" --dry-run
npx agentic-workspace-core@latest init --target "$repo_root"

Install Behavior

init is a clean core install. It may replace these managed paths:

  • AGENTS.md
  • CLAUDE.md
  • .agents/
  • docs/
  • llms.txt

Before replacement, existing agent-facing files are moved into root legacy/ with their relative paths preserved. This includes old instruction files, .agents/, docs/, llms*.txt, common AI-tool rules/config directories, and MCP config files. legacy/ is ignored by git and is not treated as active project knowledge.

This replace-first install model is deliberate: the package creates a clean active agent layer, while keeping old material available for manual review and recovery.

Update

npx agentic-workspace-core@latest update --dry-run
npx agentic-workspace-core@latest update
npm run knowledge:check

Normal update requires an existing install. It:

  • replaces AGENTS.md, CLAUDE.md, .agents/knowledge-core/, and upstream starter skill/eval files
  • preserves project docs under docs/
  • preserves project-specific skill directories and evals
  • preserves safe local .agents/knowledge.config.json extensions
  • archives obsolete agent-facing paths from older package layouts into legacy/
  • rebuilds llms.txt and .agents/generated/*
  • runs doctor validation unless --skip-check is used

Installed projects also get:

npm run awc:update:check

That command runs agentic-workspace-core@latest update --if-newer; it no-ops when the installed core is current and applies the normal update path when a newer package is available.

For controlled projects where you want the package payload reapplied from scratch:

npx agentic-workspace-core@latest update --full --dry-run
npx agentic-workspace-core@latest update --full

update --full archives the current core layer into legacy/ and reinstalls AGENTS.md, CLAUDE.md, .agents/, docs/, and llms.txt from the package payload.

What Gets Installed

AGENTS.md
CLAUDE.md
llms.txt
.agents/
  knowledge.config.json
  generated/
  knowledge-core/
  skills/
  evals/skills/
docs/
  architecture/
  components/
  decisions/
  domain/
  glossary/
  plans/
  reference/
  research/
  runbooks/
  workflows/
legacy/              # ignored, created only when old agent context is archived

.context/handoffs/ is also reserved and ignored for temporary handoff state. It is created only when a handoff workflow needs it.

Starter Skills

  • software-development-workflow: non-trivial coding work from task contract through verified completion.
  • frontend-ui-workflow: frontend/UI build, redesign, polish, review, product-appropriate design taste, design-system consistency, accessibility, responsive behavior, and visual verification.
  • project-knowledge: recall, route, write, correct, and validate durable repository knowledge.
  • research-to-knowledge: source-backed research and research persistence.
  • write-agent-skill: skill creation, revision, security review, and evals.
  • write-agent-handoff: temporary transfer state for another agent, workspace, or future session.

Project Commands

Installed projects get:

npm run knowledge:build
npm run knowledge:doctor
npm run knowledge:check
npm run awc:update:check
  • knowledge:build regenerates llms.txt and .agents/generated/*.
  • knowledge:doctor validates docs, skills, evals, config, and core paths.
  • knowledge:check verifies generated files are current and runs the doctor.
  • awc:update:check checks for a newer Agentic Workspace Core package and updates when needed.

Design Rules

  • Authored Markdown is the source of truth.
  • Generated files are navigation only.
  • Skills are procedural and do not own durable project facts.
  • Handoffs are temporary and ignored by git.
  • Memory writes require owner, evidence, scope, lifecycle status, freshness, retrieval aliases, and safety checks.
  • Optional adapters such as Obsidian, vector search, MCP, graph databases, or SQL databases must not become hidden sources of truth unless a project explicitly chooses that contract.

Requirements

  • Node.js 20 or newer.
  • npm/npx.
  • Git is strongly recommended so install and update diffs are reviewable.

Support

Open bugs, install/update problems, and package design proposals in GitHub Issues.

Package Maintainers

The publishable runtime payload lives under payload/. Source validation runs against that payload in the installed-project shape.

npm run knowledge:check
npm run smoke
npm run release:check

release:check runs knowledge validation, CLI smoke tests, and a dry-run package inspection before publishing. Releases are published to npm through GitHub Releases and the Publish workflow.

License

MIT.

Status

0.2.x is the current public hardening line. The core is usable today, with a deliberately explicit install/update model and active hardening around agent memory, skills, generated indexes, and update safety.

Reviews (0)

No results found