AI Dev Coach is a Chrome extension that helps developers use AI tools without building bad habits.
It does not block AI. It coaches behavior in real time.
AI can accelerate coding, but beginners can become dependent on copy-paste workflows before they build core engineering skills.
AI Dev Coach is designed to protect the learning loop:
- break down ambiguous problems
- explain attempts before requesting solutions
- debug with intent instead of skipping to answers
- use AI as a mentor, not a replacement for thinking
- Profile onboarding in popup (
role,level,habit goal) - First-run activation card with setup progress, trust facts, and one next best action
- Popup home actions for current prompt improvement, QA checklist, manual attempt logging, and opening a supported AI tab
- Prompt Builder with required fields (
task,context,what you tried) - In-page Quick Prompt Builder button beside AI chat composer (
Build + Insert,Build + Send) - Keyboard shortcut
Ctrl/Cmd + Owhile focused in AI chat composer (browser-dependent) - Background shortcut command
Ctrl/Cmd + Shift + Oto open Prompt Builder reliably - Developer-focused role modes: Software Engineer, Solution Architect, Engineering Manager, Other Tech Role
- Profile level dropdown: Student, Junior, Middle, Senior
- Popup workspace readiness card that explains when
InsertandSendare available - In-page Live Coach bubble with realtime prompt score and habit snapshot
- 6 prompt templates: debugging, code review, system design, refactoring, performance optimization, learning
- Prompt quality engine v2 with shared scoring across popup and live monitor (
clarity,context,specificity,risk guardrails) - Prompt Linter with rule-based feedback for prompt length, technical context, failure signals, and sensitive data
- Real-time prompt quality scoring on AI chat websites (draft + submit paths)
- Sensitive data detection and local redaction for likely secrets before prompt submission
- Warning overlays in top-right for shortcut prompts and risky copy-paste behavior
- Quiet controls on warnings and live bubble (
Snooze 1h,Less warnings,Turn off here) - AI dependency tracking (
ai requests,manual attempts,large pastes) - Daily habit progress for improved prompts, manual attempts, and QA checks
- Prompt Improvement Engine that rewrites weak drafts into structured prompts with missing context kept as placeholders
- Focused Prompt Marketplace for developer and learning prompts, with
InsertandSenddisabled until the current workspace is actually ready - Settings page for strict mode and warning thresholds
- First-run consent gate before any prompt or copied-content monitoring can be enabled
- Enterprise policy support via
chrome.storage.managed, including supported-host allowlists and admin-locked coaching settings
- ChatGPT (
chatgpt.com,chat.openai.com) - Claude (
claude.ai) - Gemini (
gemini.google.com) - Grok (
grok.com) - DeepSeek (
chat.deepseek.com)
- Clone this repository.
- Open Chrome and go to
chrome://extensions. - Enable Developer mode.
- Click Load unpacked.
- Select the
extension/directory.
node scripts/validate-extension.mjs
node scripts/test-managed-config.mjs
node scripts/test-platform-adapter.mjs
node scripts/test-prompt-quality-engine.mjs
node scripts/test-prompt-linter.mjs
node scripts/test-prompt-improvement-engine.mjs
node scripts/test-prompt-improvement-ui.mjs
find extension -type f -name '*.js' -print0 | xargs -0 -n1 node --check
python3 -m mkdocs build --strict- All documentation is Markdown under
docs/. - Docs site is built with MkDocs Material using
mkdocs.yml.
This repository includes .github/workflows/deploy-docs.yml.
On push to staging with docs-related file changes, the workflow:
- Installs
mkdocs-material - Builds docs with
mkdocs build --strict - Deploys automatically to GitHub Pages
This docs pipeline does not change extension version, tags, or releases.
- Open repository Settings > Pages.
- Set Source to GitHub Actions.
- Push to
stagingor run the workflow manually.
- Release preparation now lives in
.github/workflows/prepare-release.yml. - When extension code lands on
main, it opens or updates a release PR likechore(release): cut vX.Y.Zinstead of pushing directly to protectedmain. - The repository should allow GitHub Actions to create pull requests. If that setting is ever disabled, the workflow can use
REPO_ADMIN_TOKENso release preparation still runs without manual rescue. - After that release PR is approved and merged,
.github/workflows/release.ymlpublishes the release:- reads the manifest version from
main - creates the release tag
vX.Y.Zif missing - generates friendly, plain-language release notes
- packages the extension zip
- creates or updates the GitHub Release
- uploads and publishes the package to Chrome Web Store
- reads the manifest version from
- If Chrome Web Store reports that the item is already in review, the workflow keeps the GitHub release successful and marks the store publish as deferred instead of failing the whole run.
- Version bump is release-only on
main. PRs intostagingare guarded from release-version edits.
Configure these repository secrets before a main release run:
CWS_CLIENT_IDCWS_CLIENT_SECRETCWS_REFRESH_TOKENCWS_PUBLISHER_IDCWS_EXTENSION_ID
Optional:
CWS_PUBLISH_TARGET(defaultortrustedTesters)REPO_ADMIN_TOKEN(recommended for release PR creation, branch-protection automation, and weekly sync admin merge fallback)
The release workflow exchanges the refresh token for an access token using OAuth, then calls the Chrome Web Store API upload and publish endpoints.
stagingis the integration branch for pull requests.mainis the stable release branch.- Weekly automation syncs
stagingintomainwhen updates exist. - Branch protections enforce PR flow with 1 approval (admins can bypass for emergencies).
See docs/07-project/branching-strategy.md.
extension/
vscode-extension/
docs/
.github/workflows/ci.yml
.github/workflows/deploy-docs.yml
.github/workflows/release.yml
.github/workflows/staging-version-guard.yml
.github/workflows/weekly-staging-to-main.yml
.github/workflows/apply-branch-protection.yml
mkdocs.yml
See docs/07-project/roadmap.md.
Prompt Builder details and wireframe: docs/08-product/prompt-builder-system.md
- Analysis runs locally in browser content scripts
- No prompt or code is sent to external servers
- No remote analytics in MVP
MIT