0.3.8: Code-linked behaviour context

Version 0.3.8 helps you change code without losing sight of why it was written. Every file in a long-running project carries promises: what it was meant to do, what someone agreed it would handle and which tests were meant to prove it. Those promises usually sit in old plans that nobody rereads. EWAI now brings them back at the moment they matter.

Ask a file what it's responsible for. Before you change a file, run ewai index behaviour <path> or simply ask your assistant. You'll see the outcomes earlier work linked to it, the acceptance criteria behind them, and the tests and evidence recorded when they were delivered, with the most relevant first.

Know when old evidence has gone stale. EWAI checks each piece of recorded evidence against what's on disk now and tells you if it's changed. It's honest about what it can't see, too. Finding nothing doesn't mean nothing is affected, and EWAI won't claim old work is still correct. That judgement stays with you.

Your assistant remembers too. When you plan, build or review, EWAI adds a short reminder of earlier outcomes linked to the files you're about to touch, so your assistant can recheck them without being asked. It's kept brief, so the work in front of you still comes first.

Confirm what you actually reviewed. Impact assessments now show the earlier outcomes worth rechecking. If anything changes between your review and your confirmation, EWAI asks you to take another look rather than recording a decision on evidence you never saw. If you confirm assessments through the API, include the previewId from the preview.

The Institute and the free starter persona pack

Version 0.3.7 celebrates more than 2,000 developers working with Engineering With AI. Thank you. To mark it, we've added something we think you'll enjoy: the Institute.

Watch your delivery come to life. The Institute is a living cutaway of your work. Open it from the dashboard sidebar and every intent appears in the room for its stage, with the assistant working on it and the personas you've attached walking about alongside. You can see at a glance where work is moving, where it's waiting and who's involved, without opening a single card. Zoom into any room for a closer look, then press Esc or choose ← DASHBOARD to return to the normal view.

Know who's waiting for you. When work needs a decision from you, its people wave for your attention and a thermometer fills the longer they wait. The same "waiting since" time now shows on your board too.

Try personas for free. The new free starter pack gives you ten specialist personas, from an architect to a CISO, to try inside EWAI. Claim your free starter pack, enter your key and they install in one step. When you're ready for the full library, upgrading replaces the starter pack cleanly.

0.3.6: Repository context

Version 0.3.6 adds repository context. Your assistant can read a short, evidence-based summary of the file in front of it: the callers EWAI resolved, recorded routes, who usually changes it, which files change with it and which project records name it. A change rank orders your current diff against the repository's own recent commits, so the riskiest edits are looked at first. The dashboard's Repository view shows the same readings. They are advisory and never approve Build, Manual QA or release.

Keep the index current without rebuilding it. ewai index update re-parses only the files that were added, changed or deleted, and a Git post-commit hook runs it in the background. A hook failure never undoes your commit. Resolution now follows relative imports, Composer PSR-4 prefixes, TypeScript and JavaScript path maps, package imports, Go module prefixes and common aliases, and it recognises common route conventions.

Shorter test and Git output. ewai distill -- <command> keeps a short reading of recognised test and Git commands with the exit code intact, and stores the full log so you can hash it as gate evidence. Only a single plain command is shortened. Chained, piped or redirected commands are left alone, and your assistant's normal approval rules still apply.

Hooks EWAI adds to a project. When EWAI writes its assistant configuration for a project, it also adds Claude Code hooks for session, file and command readings, plus the Git post-commit hook. ewai context install --project . applies them again, and setting EWAI_DISTILL=0 turns command shortening off. The repository context guide in the documentation describes each reading and its limits.

0.3.3: Clean package homepage link

Package metadata only; no runtime changes. The npm package homepage now uses the short address conversationalcoding.dev/ewai.

0.3.2: Package links and quick start

Documentation and package metadata only; no runtime changes.

• The npm package homepage now points to the Engineering With AI website.
• The README adds a product screenshot, links to the website, online documentation and persona library, and a short quick start listing the supported assistants: Claude Code, Codex, Google Antigravity and Grok Build.
• Broader package keywords make the package easier to find on npm.

Update with npm install --save-dev @thebackstoryis/engineering-with-ai. The 0.3.1 features (concise guidance and Grok Build) are unchanged.

0.3.1: Concise guidance and Grok Build

Version 0.3.1 brings concise communication guidance and Grok Build support to the stable npm channel.

Concise answers and guided decisions

• EWAI's managed project instructions prioritise correctness and usefulness, then brevity: the result or recommendation first, with less repetition and routine narration.
• Decision requests explain the options and consequences, then give a recommendation with its reason and a suggested response when useful.
• Unclear requests are refined with a recommended interpretation and the assumptions that matter.
• The guidance is designed to reduce unnecessary output. Tool-result compaction is still planned, and token or cost savings haven't been measured yet.

Grok Build and coding provider settings

• Grok Build joins Codex, Claude Code and Google Antigravity as a supported host, with native companion launch, skills and project MCP setup.
• Choose a primary coding CLI, independent reviewers and an eligible pool for unattended work in Configuration → Coding providers or with ewai providers. Each CLI keeps its own model choice by default.
• Grok can run isolated coding, read-only review and restricted proposal workers. Each mode passes an offline conformance check before dispatch.
• Save an xAI key privately, outside the project, through the dashboard or a hidden terminal prompt.

Build approval, required reviews and Manual QA are unchanged.

Update

npm install --save-dev @thebackstoryis/engineering-with-ai
npx ewai

Initialisation or the next check-in refreshes the managed block in AGENTS.md and CLAUDE.md, preserving project-authored guidance outside it. Start a fresh host conversation afterwards.

Validation: the full suite passed (889 passed, 10 skipped) with publication boundary, syntax and focused Team Hub coverage checks. The package is published with npm provenance.

0.3.1-beta.0: Concise guidance and query refinement

Version 0.3.1-beta.0 adds concise communication guidance to EWAI's managed project instructions.

• Prioritise correctness, fidelity and useful action details while reducing repetition and routine narration.
• Support decisions with relevant options, consequences, a recommendation and suggested responses.
• Refine unclear requests by leading with the recommended interpretation and consequential assumptions.
• Include user guidance for installation, existing-project refresh and returning to production.

Initialisation or normal check-in updates the managed block in AGENTS.md and CLAUDE.md while preserving project-authored guidance outside it. Start a fresh host conversation after the refresh.

Install the beta:

npm install --save-dev @thebackstoryis/engineering-with-ai@beta
npx ewai

Tool-result compaction remains planned. Incremental token or cost savings have not been measured. Mandatory workflow and human approvals continue to apply. Production remains on 0.3.0.

Validation: syntax checks and publication-boundary checks passed; the full suite passed 838 tests with one skipped; focused Team Hub coverage passed 28 tests. A clean installation of the packed beta verified initialisation, instruction refresh, local-guidance preservation and packaged user documentation.