Guided Audit Conductor - Audity for Agents

What the Guided Audit Conductor is

The Audity MCP server ships an instructions brain that turns any connected, instruction-aware agent (Claude, ChatGPT, Cursor) into a guided conductor. It automatically runs the audit pipeline in order, previews cost and time before any paid step, polls long jobs correctly, and flags confidence thresholds. The conductor handles the orchestration details so you can focus on the workflow prompt.

Pipeline order

Every audit follows this sequence:

Intake (create project, upload documents, create interview sessions)
Document analysis (run audity_enqueue_document_analysis, poll audity_get_job)
Interview analysis (run audity_enqueue_interview_analysis, poll audity_get_job)
Comprehensive audit (requires BOTH analyses; run audity_enqueue_audit_analysis, poll audity_get_job)
Opportunities + ROI (run audity_select_opportunities, audity_create_roi_calculation)
Deliverables (PDF, Gamma deck, stakeholder memos; run in parallel: audity_enqueue_pdf_generation, audity_enqueue_gamma_deck, audity_enqueue_stakeholder_memos)

The conductor knows this order and enforces it automatically.

Confirm-before-spend

Before running any paid or generation step (audit analysis, deliverable generation), the conductor:

Calls audity_get_navigation_status to check prerequisites
Previews the cost (e.g. “~1,000 credits”) and typical time (a comprehensive audit usually runs ~10-15 minutes; lighter steps a few minutes)
Waits for your approval before proceeding
Continues only after you confirm

This prevents accidental credit overspends and gives you a checkpoint before long operations.

Long-job polling

The MCP server ships with built-in polling for async jobs:

audity_get_job returns { status: "pending" | "completed" | "failed", ... }
The conductor polls automatically and knows typical timing per job type: a comprehensive audit usually runs ~10-15 minutes, document and interview analyses a few minutes each, and deliverable jobs return quickly (the file then arrives via callback, see below)

Important: A finished Gamma deck or PDF job means generation was TRIGGERED, not that the file is ready to download. After generation completes, poll audity_get_gamma_status or audity_get_pdf_status to confirm the file is ready.

Grounding and confidence

Findings returned by the API include a confidence field (0.0–1.0, typically 0.6–1.0). The conductor flags any finding under 0.8 confidence as a hypothesis rather than a fact:

“Potential opportunity (confidence 0.72): implement RPA for invoice processing. Verify with the client before committing to this initiative.”

This reminds you that AI-generated patterns are educated guesses, not assertions.

Using the conductor in Claude Code

Claude Code supports a /audit skill that triggers conductor mode. connect offers to install it while you run npx @auditynow/connect (default yes when Claude Code is detected), there is no separate flag. Then in Claude Code:

/audit Start an Audity project for "Acme Corp", mid-market manufacturing. 
Run a comprehensive audit. Summarize the top three opportunities.

The /audit skill automatically:

Runs the pipeline in order
Asks for approval before spend
Polls jobs correctly
Flags confidence thresholds
Reports findings with IDs so you can drill in

Common prompts (all agents)

Start a new audit for {client}, {industry}, {size}. Run the full analysis. 
Give me the top three opportunities ranked by impact.

Run audit analysis on project {id}. When it's done, pull the opportunities,
rank by ROI potential, and tell me which three are most realistic given the 
client's current IT maturity.

Generate the PDF and Gamma deck deliverables for project {id}. 
Confirm cost before proceeding.

What patterns am I seeing across my healthcare clients? 
Pull opportunities from the last three audits, tag by pattern, confidence.

How it differs from direct API calls

Direct API call	Conductor
You manually call `audity_enqueue_audit_analysis`, then `audity_get_job` in a polling loop	Conductor enqueues and polls automatically
You verify prerequisites manually (e.g. checking both analyses exist)	Conductor checks prerequisites and gates steps
You estimate cost; each `POST /api/projects` deducts 1,000 credits	Conductor previews cost before you approve
Confidence fields are raw numbers; you interpret them	Conductor flags under-0.8 confidence as hypotheses
You reconstruct the “next step” after each completion	Conductor knows the pipeline order and enforces it

For simple read-only queries (“list my projects”, “fetch opportunities for project X”), direct API calls are fine. For orchestrated workflows (full audits, batch lead conversion, deliverable generation), the conductor saves orchestration logic.

Troubleshooting

Conductor doesn't start or hangs

Ensure you ran npx @auditynow/connect and that your tool restarted (especially Claude Code, which loads its MCP servers only at startup). If using Claude Desktop, go to Settings → Connectors and confirm Audity is green (connected).

Conductor says a prerequisite isn't met

The audit pipeline requires both document and interview analyses before running comprehensive audit. If you skipped one, run it via audity_enqueue_document_analysis or audity_enqueue_interview_analysis first, then retry comprehensive audit.

Job polling times out or gets stuck

A comprehensive audit can take ~10-15 minutes (sometimes longer); lighter jobs finish in a few minutes. If a job hasn’t progressed well past its typical time, pull the job details via audity_get_job and check for an error message. Restarting the conductor will pick up where it left off.

PDF or Gamma generation says 'triggered' not 'ready'

Correct behavior. Generation jobs return { status: "completed" } when TRIGGERED, not when the file is fully ready. Poll audity_get_pdf_status or audity_get_gamma_status after generation completes to confirm the file is ready for download.

What’s next

Run a full audit workflow →, detailed end-to-end recipe (without the conductor)
Connect with one command →, npx @auditynow/connect setup
Lead conversion playbook →, convert ReadyLink leads from your agent

​What the Guided Audit Conductor is

​Pipeline order

​Confirm-before-spend

​Long-job polling

​Grounding and confidence

​Using the conductor in Claude Code

​Common prompts (all agents)

​How it differs from direct API calls

​Troubleshooting

​What’s next