@shree_builds

Devlog #4

We can finally see some success in actually reducing token count significantly. PromptGuard has been able to achieve significant token reduction in prompts containing system logs. We have documented a ~64% token reduction in that area of prompting.

There are now 3 other types we need to find ways to reduce token counts for. They are code, text and JSON data.

A few new updates have been made to PromptGuard…
PromptGuard is not only a VS Code extension now; it is a token-reducing proxy as well. The PromptGuard-Extension now can only show dashboard metrics fetched from the proxy and expand prompts if the prompt is too vague. The PromptGuard-Proxy runs locally on port 8787 by default and carries out all the token reduction and cost estimate required by the dashboard. PromptGuard is not merely an extension anymore, it is an ecosystem.

I will be publishing the Python package for the proxy by Monday(IST) evening. You can view the Proxy’s source here: https://github.com/ShreeyashBihade/PromptGuard-Proxy

Devlog #4

We can finally see some success in actually reducing token count significantly. PromptGuard has been able to achieve significant token reduction in prompts containing system logs. We have documented a ~64% token reduction in that area of prompting.

There are now 3 other types we need to find ways to reduce token counts for. They are code, text and JSON data.

A few new updates have been made to PromptGuard…
PromptGuard is not only a VS Code extension now; it is a token-reducing proxy as well. The PromptGuard-Extension now can only show dashboard metrics fetched from the proxy and expand prompts if the prompt is too vague. The PromptGuard-Proxy runs locally on port 8787 by default and carries out all the token reduction and cost estimate required by the dashboard. PromptGuard is not merely an extension anymore, it is an ecosystem.

I will be publishing the Python package for the proxy by Monday(IST) evening. You can view the Proxy’s source here: https://github.com/ShreeyashBihade/PromptGuard-Proxy

Devlog #3

SummaryContinued from the previous devlog by finishing the core migration hardening, validating extension compatibility, and completing the final product-surface cleanup for the retired features.

What Was Completed Since Last Devlog:
Finished stage-4 governance extraction inside core by stabilizing service imports and helper dependencies. Added missing core helper modules needed by governance services.

Normalized and fixed core service pathing and module export issues so core builds cleanly.

Finalized core exports in index.ts for migrated modules.

Wired workspace/package metadata for core package usage

Completed final cleanup pass for retired feature surfaces (UI/service/test remnants removed from active extension codebase).

Couldn’t write much today, but you can now clone the github repository and test PromptGuard in VSCode on your own.

2 Unique Selling Points:

1. PromptGuard can elaborate a short or less expressive prompt tailored to your needs by asking helper questions. (Requires a Groq API key in the project workspace)

2. PromptGuard can minimize 1 to 400 tokens on prompts exceeding 350 tokens.

PS: The Groq API key is free, so please get it to enjoy all the features of PromptGuard. You won’t see cost savings as of now.

Devlog #3

SummaryContinued from the previous devlog by finishing the core migration hardening, validating extension compatibility, and completing the final product-surface cleanup for the retired features.

What Was Completed Since Last Devlog:
Finished stage-4 governance extraction inside core by stabilizing service imports and helper dependencies. Added missing core helper modules needed by governance services.

Normalized and fixed core service pathing and module export issues so core builds cleanly.

Finalized core exports in index.ts for migrated modules.

Wired workspace/package metadata for core package usage

Completed final cleanup pass for retired feature surfaces (UI/service/test remnants removed from active extension codebase).

Couldn’t write much today, but you can now clone the github repository and test PromptGuard in VSCode on your own.

2 Unique Selling Points:

1. PromptGuard can elaborate a short or less expressive prompt tailored to your needs by asking helper questions. (Requires a Groq API key in the project workspace)

2. PromptGuard can minimize 1 to 400 tokens on prompts exceeding 350 tokens.

PS: The Groq API key is free, so please get it to enjoy all the features of PromptGuard. You won’t see cost savings as of now.

Today’s session focused on stabilising the end-to-end prompt workflow, enforcing onboarding rules, improving UX, and shipping better local analytics/control features.

Onboarding and auth flow hardening Enforced onboarding as the first gate for cloud-assisted analysis.

Fixed onboarding sequencing to email -> OTP -> project name. Prevented repeated onboarding when a valid token/session already exists. Resolved auth race issues by persisting session at the right point. Added logout and delete-account actions. Analysis path controls (local vs. Groq) Added per-prompt path selection so users can choose a local or Groq-assessed flow each run. Added persistent mode toggle options: always ask, prefer local, prefer Groq. Added one-time project-start preference prompt plus /settings and /preferences entry points. Added fallback behaviour so users can continue with local analysis even if onboarding is refused. Prompt processing + persistence Wired pipeline so prompts can be saved to the backend when the cloud path is authorised.

Added local heuristic insights and model recommendations in local mode. Improved consistency of score/trace behaviour across local and Groq paths. Dashboard and chat UX upgrades Improved onboarding screen polish and overall flow clarity. Added inline expand/minimise and a cleaner action layout in chat/dashboard interactions. Kept control buttons on one line and improved visual hierarchy.

Added motivational summary visuals and better history presentation. Local optimization ledger + backups Implemented local JSON ledger for input/output prompts, token counts, and savings tracking. Added rolling backups before overwrite and cleanup logic for old backups. Included clear/reset behaviour tied to account/data controls. Optimization comparison experience Replaced temporary markdown diff flow with a dedicated comparison panel. Added a side-by-side original vs. optimised prompt view. Added the Copy Optimised Prompt action directly in the comparison UI.

Validation: Build and test checks passed after the latest implementation round. Core features now operate with better resilience in both cloud-assisted and local-only scenarios.

Net result: PromptGuard is now significantly more reliable, more user-controlled, and more production-ready in onboarding, analysis routing, optimisation review, and local telemetry.

Today’s session focused on stabilising the end-to-end prompt workflow, enforcing onboarding rules, improving UX, and shipping better local analytics/control features.

Onboarding and auth flow hardening Enforced onboarding as the first gate for cloud-assisted analysis.

Fixed onboarding sequencing to email -> OTP -> project name. Prevented repeated onboarding when a valid token/session already exists. Resolved auth race issues by persisting session at the right point. Added logout and delete-account actions. Analysis path controls (local vs. Groq) Added per-prompt path selection so users can choose a local or Groq-assessed flow each run. Added persistent mode toggle options: always ask, prefer local, prefer Groq. Added one-time project-start preference prompt plus /settings and /preferences entry points. Added fallback behaviour so users can continue with local analysis even if onboarding is refused. Prompt processing + persistence Wired pipeline so prompts can be saved to the backend when the cloud path is authorised.

Added local heuristic insights and model recommendations in local mode. Improved consistency of score/trace behaviour across local and Groq paths. Dashboard and chat UX upgrades Improved onboarding screen polish and overall flow clarity. Added inline expand/minimise and a cleaner action layout in chat/dashboard interactions. Kept control buttons on one line and improved visual hierarchy.

Added motivational summary visuals and better history presentation. Local optimization ledger + backups Implemented local JSON ledger for input/output prompts, token counts, and savings tracking. Added rolling backups before overwrite and cleanup logic for old backups. Included clear/reset behaviour tied to account/data controls. Optimization comparison experience Replaced temporary markdown diff flow with a dedicated comparison panel. Added a side-by-side original vs. optimised prompt view. Added the Copy Optimised Prompt action directly in the comparison UI.

Validation: Build and test checks passed after the latest implementation round. Core features now operate with better resilience in both cloud-assisted and local-only scenarios.

Net result: PromptGuard is now significantly more reliable, more user-controlled, and more production-ready in onboarding, analysis routing, optimisation review, and local telemetry.

PromptGuard — Devlog #1

Today we built the first working version of PromptGuard: a local-first VS Code extension that helps people write safer, clearer, and cheaper AI prompts.The core focus was making prompt review useful without sending prompts anywhere by default.

PromptGuard now analyzes prompts locally for ambiguity, missing roles/constraints/output formats, repetition, prompt-injection patterns, potential secrets, and PII. It turns those checks into a quality score, actionable findings, editor decorations, and quick fixes.We also added practical prompt-engineering tooling: token, cost, latency, and savings estimates; locally stored searchable prompt history; an analytics dashboard; a sidebar navigator; and a non-destructive optimization diff so users can inspect a suggested rewrite before adopting it.

For VS Code Chat, we added @promptguard /analyze, which reports a prompt’s score, safety findings, and token information directly in chat. The extension respects provider boundaries: it only handles chat turns sent to PromptGuard itself, rather than attempting to intercept other providers’ messages.

We added optional Groq-powered semantic review and prompt improvement too. It is strictly opt-in through a local API key, clearly warns that prompts leave VS Code when enabled, asks targeted follow-up questions when key context is missing, and validates the improved prompt locally afterward.

Finally, we set up the extension project structure, local development configuration, icon, documentation, licensing, and a modular architecture for future rule plugins and model providers. The project type-checks and compiles successfully.

Next up: add dashboard and diff screenshots, benchmark prompts, and build toward policy packs and CI prompt linting.

PromptGuard — Devlog #1

Today we built the first working version of PromptGuard: a local-first VS Code extension that helps people write safer, clearer, and cheaper AI prompts.The core focus was making prompt review useful without sending prompts anywhere by default.

PromptGuard now analyzes prompts locally for ambiguity, missing roles/constraints/output formats, repetition, prompt-injection patterns, potential secrets, and PII. It turns those checks into a quality score, actionable findings, editor decorations, and quick fixes.We also added practical prompt-engineering tooling: token, cost, latency, and savings estimates; locally stored searchable prompt history; an analytics dashboard; a sidebar navigator; and a non-destructive optimization diff so users can inspect a suggested rewrite before adopting it.

For VS Code Chat, we added @promptguard /analyze, which reports a prompt’s score, safety findings, and token information directly in chat. The extension respects provider boundaries: it only handles chat turns sent to PromptGuard itself, rather than attempting to intercept other providers’ messages.

We added optional Groq-powered semantic review and prompt improvement too. It is strictly opt-in through a local API key, clearly warns that prompts leave VS Code when enabled, asks targeted follow-up questions when key context is missing, and validates the improved prompt locally afterward.

Finally, we set up the extension project structure, local development configuration, icon, documentation, licensing, and a modular architecture for future rule plugins and model providers. The project type-checks and compiles successfully.

Next up: add dashboard and diff screenshots, benchmark prompts, and build toward policy packs and CI prompt linting.