Help manual

How to use ATower Of Agents.

A practical operating guide for creating workflows, preparing evidence, using the Band agent room, and understanding current MVP limits.

Create workflow

Quick Start

  1. Open Workflows and create an HR Candidate Screening draft.
  2. Upload the candidate resume, job description, and hiring policy.
  3. Use the Band room to mention the coordinator and specialists.
  4. Review specialist findings as advisory output before any human decision.

Band agents are infrastructure. Users should not create remote agents or handle API keys per workflow. They create workflows, upload evidence, and use the configured room.

Workflow Types

Flagship MVP path

HR Candidate Screening

Collects resume, job description, and hiring policy evidence so specialists can compare requirements, check policy, audit bias risk, plan interviews, and summarize advisory findings.

Template scaffold

Sales Lead Qualification

Uses supplied ICP, CRM notes, and sales context to reason about lead fit and next-step recommendations without inventing prospect facts.

Template scaffold

Engineering Change Review

Uses supplied design notes, code changes, and test evidence to review implementation risk, missing tests, and integration concerns.

Evidence Required

WorkflowRequired artifacts
HR Candidate ScreeningCandidate resume, job description, hiring policy
Sales Lead QualificationICP, CRM notes, prospect context, relevant sales policy
Engineering Change ReviewDesign notes, code/change summary, test evidence, rollout constraints

Band Room Guide

One-time setup versus daily use

Admins create the Band remote agents once and store credentials in environment variables. End users only create workflow drafts, upload artifacts, and mention agents in the configured room.

  1. Open the shared ATower Band room.
  2. Confirm the coordinator and specialists are participants.
  3. Mention agents with their full handle, including the user namespace.
  4. Never paste credentials, resumes, or private docs into public rooms.

Agent Roster

@hilaytrivedi1224/atower-coordinator

Primary room coordinator. Routes work, explains required evidence, and keeps the room honest about current product limits.

@hilaytrivedi1224/atower-rag-retriever

Checks evidence availability and retrieval status. It must not invent evidence IDs or citations.

@hilaytrivedi1224/atower-policy-guardian

Reviews supplied policy and compliance evidence, then flags conflicts, missing approvals, and escalation needs.

@hilaytrivedi1224/atower-final-decision

Synthesizes advisory findings for human review. It does not make autonomous high-impact decisions.

@hilaytrivedi1224/atower-resume-jd-matcher

Compares resume evidence against job-description requirements and reports matches, gaps, and uncertainty.

@hilaytrivedi1224/atower-bias-reviewer

Audits screening reasoning for unfair assumptions, proxy risks, unsupported inferences, and biased language.

@hilaytrivedi1224/atower-interview-planner

Creates fair, job-related interview follow-up questions from evidence gaps.

@hilaytrivedi1224/atower-lead-qualifier

Reviews supplied ICP, CRM, and sales-note evidence for lead-fit findings and follow-up recommendations.

@hilaytrivedi1224/atower-engineering-reviewer

Reviews supplied engineering changes, design notes, and tests for implementation risk and missing coverage.

Copyable Prompts

Coordinator smoke test

@hilaytrivedi1224/atower-coordinator confirm you are connected, summarize the available specialist roster, and state what evidence is required before running HR candidate screening.

HR workflow kickoff

@hilaytrivedi1224/atower-coordinator We are preparing an HR Candidate Screening workflow in ATower. The required artifacts are resume, job description, and hiring policy. Please coordinate the available specialists and explain what each agent should review before a human makes any screening decision.

Specialist test

@hilaytrivedi1224/atower-resume-jd-matcher confirm you are connected and explain what evidence you need before matching a resume to a job description.

Current MVP Limits

  • Workflow execution is not implemented yet; run endpoints return 501.
  • Agent outputs are advisory and require human review.
  • Agents must not invent evidence IDs, citations, scores, or outcomes.
  • Missing provider credentials must produce explicit unconfigured behavior.
  • Band messages are the visible audit layer, not a substitute for persisted findings.

Troubleshooting

No agent reply

Confirm the agent is a room participant, was mentioned with its full handle, and the band-agent service is running.

Provider errors

Check LLM_PROVIDER, provider API keys, and that the selected model supports OpenAI tool calling.

No workflows listed

Set NEXT_PUBLIC_DEFAULT_ORG_ID and confirm the API is reachable at NEXT_PUBLIC_API_BASE_URL.