Inspiration
A difficult workplace conversation should not be a manager's first rehearsal. Manager Mirror is a focused practice room for trying an opening, seeing how it could land, and leaving with clearer words and a concrete next step. It uses fictional situations rather than employer records or employee profiles.
Current prototype
The current web prototype contains four fictional scenarios: a missed commitment, an after-hours boundary reset, saying no to additional scope, and team conflict. A user writes a response and requests a counterpart reply, a six-dimension coaching scorecard, strengths and risks, an improved opening, and a follow-up draft.
The dimensions are directness, empathy, curiosity, accountability, next step, and composure. They are coaching heuristics, not validated assessments of an employee or a manager's performance.
Agent implementation and verification
The AppDeploy source includes an integration intended to invoke two Strands Agents: a fictional counterpart and a coaching agent. Its model adapter calls the hosting platform's AI service. The implementation also contains a deterministic fallback rubric and output validation. Deployment availability alone does not establish that a live Strands run succeeded; that end-to-end behavior still requires verification before final competition submission.
A separate Replit project is being repaired and tested. Its update request has been accepted, but the completed update, a public source repository, and a demonstration video have not yet been verified.
Privacy and action boundary
The prototype uses seeded fictional cases and screens input for common contact details, credentials and confidential markers. This screening is limited and is not a guarantee that every sensitive detail can be detected; users must not enter private workplace information. The app drafts text only. It does not contact anyone, send follow-ups, invite teammates or testers, request referrals, access an employee directory, or take employment action.
It is a practice tool, not legal, HR, medical, or employment advice.
Intended competition
The intended target is Agents for Humans, Professional Agents. This page is a project portfolio, not evidence of a finalized entry in that hackathon. A complete submission needs verified Strands execution, a public MIT- or Apache-licensed repository, an architecture diagram, an AWS Builder ID and a qualifying demonstration video.
The existing Agentic Cinema association should not be treated as a valid entry: this management-coaching prototype does not currently demonstrate that competition's media-and-entertainment workflow requirements.
Planned, not yet claimed
Native mobile publication, voice practice, RevenueCat purchases, reminders, private persistent history, OneSignal, Layers, and measured customer outcomes are planned capabilities, not verified features of this prototype. No revenue, user count, award, or completed submission is claimed.
What is next
Verify the end-to-end agent workflow and failure handling, publish reproducible licensed source, record an accurate working demonstration, attach the required architecture diagram, and finalize the correctly targeted entry only with truthful evidence.
Built With
- appdeploy-ai
- node.js
- react
- replit
- strands-agents-sdk
- typescript
- vitest
- zod