Feature Specification¶

1. Developer Profile¶

Users can store:

job role
level (Student, Junior, Middle, Senior)
habit goal

Job role options:

Teacher
Software Engineer
Solution Architecture
Manager
Director
Doctor
Other

This profile personalizes coaching tips.

2. Prompt Builder¶

Templates are provided for:

debugging
code review
system design
refactoring
performance optimization
learning

Required fields:

task
context
what user already tried

The extension generates a structured prompt and provides copy-to-clipboard.

In-page access:

floating launcher beside composer on supported AI pages
keyboard shortcut Ctrl/Cmd + O to open Prompt Builder when chat input is focused (browser-dependent)
command shortcut Ctrl/Cmd + Shift + O to open Prompt Builder reliably

2.1 Prompt Marketplace¶

The popup now includes a local Prompt Marketplace powered by a markdown prompt library.

Capabilities:

browse prompt cards by category
search prompt text locally
filter by category
show inline suggestions in the active AI chat composer while the user types
copy a prompt
insert a prompt into the active supported AI chat
insert and auto-send a prompt into the active supported AI chat

Marketplace data flow:

source markdown lives in docs/prompts.md
a sync script exports that markdown into the extension bundle
the extension parses markdown into structured prompt objects locally
local usage counts drive trending prompts
inline suggestion ranking reuses local prompt usage, role weighting, and query intent matching

Prompt categories:

Developer
Learning
Writing
Productivity
Image Generation
Business
Daily Life

3. Prompt Quality Analyzer¶

Analyzes each submitted prompt for:

task clarity and vague-language detection
context evidence completeness (error/failure + expected/actual + artifact)
framework or stack context
independent attempt quality (action + result + blocker)
negative attempt phrases (for example, \"I didn't try\", \"chua thu\") as penalties
shortcut intent signals

Outputs a shared quality score and targeted feedback across both popup prompt builder and live monitoring.

Role-aware prompt feedback now adds:

role-specific warning copy
role-specific evidence suggestions
software-engineering specialization hints for frontend, backend, DevOps, and fullstack wording
level-aware nudges for Student and Junior

3.1 Prompt Linter¶

The extension also runs a lint-style ruleset to make feedback easier to scan before sending.

Initial lint rules:

prompt too short
missing technical context
missing concrete error message for failure-oriented prompts
possible sensitive data

Lint output:

shows pass and fail states in popup prompt builder
reuses the same rules in live monitoring
only blocks normal prompting when a severe rule is triggered, such as sensitive data detection

Trigger behavior:

run in draft mode while typing with 500ms debounce (live bubble updates)
run immediately when user presses Enter
run immediately when user clicks AI platform send/submit button

4. Habit and Dependency Coaching¶

Tracks local counters:

AI requests
manual attempts
large code pastes
long AI output copies
fast copies (copied too soon after AI response)

Displays warning when dependency exceeds configured threshold.

Live bubble behavior:

bubble stays visible on supported AI pages and auto-remounts if site UI rerenders
bubble updates from both draft typing analysis and send-time analysis

4.1 Learning Analytics (V2 Foundation)¶

The first V2 analytics slice tracks prompt-session metadata locally and builds a daily summary from that history.

Tracked prompt-event fields currently include:

prompt source (composer_submit, quick_builder, send-only variants)
platform
prompt category
timestamp
prompt length
prompt score and grade
dependency percentage at send time
independent-attempt flag
shortcut-intent flag
warning count
lint failed count
role and level metadata

Privacy defaults:

analytics history is stored locally in chrome.storage.local
prompt text is not stored in the analytics event log
future backend sync is documented but not enabled yet

Daily summary output includes:

total prompts for today
average quality score
top prompt categories
readable coaching suggestions

Those coaching suggestions can now reflect the latest tracked role and level metadata.

Trend dashboard output includes:

7-day quality trend
warning frequency trend
prompt category mix view
explainable calculation rules shown in the UI copy/docs

5. Sensitive Data Guardrail¶

Before a prompt is sent, the extension can scan for likely secrets such as:

AWS access keys
JWT tokens
private keys
database URLs
API keys

If sensitive content is detected:

the user gets a top-right warning before send
the prompt can be redacted locally in the chat input
the user is asked to review the prompt before sending again

All secret scanning and redaction runs locally in the browser extension.

6. Copy-Paste Risk Alerts¶

When large multi-line code pastes are detected, the user gets coaching prompts to review and test before using pasted code.

If a user copies long AI-generated output and pastes it back into a prompt composer, the extension warns them to rewrite with their own reasoning before sending.

The extension also tracks AI output timestamp and copy timestamp. If copy happens faster than configured minimum read time, it triggers a \"read first\" warning.

7. Settings and Controls¶

User can configure:

enable/disable coaching
enable/disable prompt listener
enable/disable behavior monitor
allow/disallow prompt content reading
allow/disallow copied AI content reading
strict mode
dependency warning threshold
large paste threshold
long AI copy threshold
minimum read time before copy
enable/disable output countdown reminders
overlay duration

Notification behavior:

coaching toasts render in top-right
default toast duration is longer than baseline and can be tuned in settings