Everything Agentic Engineering v0.1.0

Everything Agentic Engineering v0.1.0

v0.1.0 is the first public preview of a production-minded operating system
for humans and AI coding agents shipping software together.

It is not a prompt collection. It is a repository-centered harness that makes
product intent, durable context, agent ownership, product design, security,
verification, and GitHub collaboration explicit.

Highlights

• Durable memory: product, design, architecture, execution, evaluation, and
  handoff context live in version-controlled artifacts.
• Product-design engine: work moves through only the discovery, strategy,
  interaction, system, token, implementation, audit, critique, and polish
  phases the current task actually needs.
• Project-specific taste: references are ingredients, components are
  structural donors, tokens encode approved decisions, and the project design
  system wins.
• Safe capability profiles: web, mobile, backend, research, and design
  capabilities are selected explicitly. Inactive capabilities are retained but
  not routed or treated as requirements.
• Task execution launcher: TASKS.jsonl entries become reviewed branch or
  worktree plans with dependency, ownership, profile, agent, and verification
  gates.
• Evidence over confidence: builders do not certify their own substantial
  UI work; running-product evidence and an independent evaluator do.
• Human-gated improvement: daily research can propose ecosystem updates,
  but adoption, commits, merges, releases, and deployment remain human actions.
• Signalroom: a complete responsive reference experience demonstrating
  agent progress, approvals, interruption, recovery, and durable artifacts.

Try it

git clone https://github.com/Gaurav890/everything-agentic-engineering.git
cd everything-agentic-engineering
cp .env.example .env
./scripts/bootstrap.sh
./scripts/init-project.sh
./scripts/profile-doctor.sh

Then create a real requirement and task, inspect the plan, and explicitly start
its isolated workspace:

./scripts/task-plan.sh T-101
./scripts/task-start.sh T-101 --yes

Supported baseline

• macOS and Linux / GitHub Actions
• Windows through WSL on a best-effort basis
• Claude Code natively
• Codex through AGENTS.md and shared repository contracts
• Other coding agents when they honor the source-of-truth and safety model

Known limitations

• Signalroom is a reference product, not a connected agent runtime.
• Profiles route and validate capabilities but do not install or delete tools.
• Optional external skills and MCPs require separate setup.
• Native Windows wrappers are not included.
• Repository settings, public releases, deployments, and announcements remain
  human-controlled.
• The self-improvement loop proposes changes; it cannot adopt them itself.

See Compatibility for the maintained matrix.

Upgrade notes

This is the first tagged preview, so there is no prior tagged migration. Before
customizing the starter, keep project-specific changes on short-lived branches
and preserve the durable artifact contracts so later upgrades can be reviewed
rather than blindly copied.

Verification

The release candidate must pass:

• a clean-checkout onboarding smoke test;
• repository JSON, JSONL, YAML, shell, Python, skill, link, token, profile,
  initializer, task-engine, hook, and evidence validation;
• Signalroom build, interaction, responsive, and accessibility checks when its
  files are in scope;
• human review of the release notes, limitations, and public README.

Publishing v0.1.0 is intentionally a separate, explicit maintainer action.