| name | client-audit |
| description | Turn raw client conversations and your honest read of a business into a professional, ROI-ranked audit and execution roadmap. Use when the user wants to create a client audit, business audit, ROI-ranked roadmap, or audit report from call notes, transcripts, or conversation dumps. Triggers: client audit, business audit, run an audit, audit this client, ROI roadmap. |
Client Audit Skill
An audit is not a writing task. It's a judgement task.
Any model can format a document. Clients pay for the judgement: what's broken, what it's costing them, and what to fix first. This skill turns raw material — conversations, notes, your honest read — into an audit that ranks every problem by ROI and hands back a roadmap.
Best run on the strongest model available (judgement-heavy). The output roadmap is designed to be executed by cheaper models over months. Judge once, execute forever.
Inputs
Ask for (or locate in the working directory) three things:
- Raw conversations — call transcripts, WhatsApp/email threads, meeting notes with the client. Unedited. More is better.
- The operator's raw read — the user's honest, unpolished take on the business: what they suspect is broken, what the client won't admit, gut feel included. Never skip this. The audit's judgement should align with and sharpen the operator's judgement, not replace it.
- Business context — industry, team size, rough revenue if known, what the client already tried.
If only conversations are provided, ask for the raw read before writing anything. It is the highest-signal input.
Process
Step 1 — Extract problems, not topics
Read everything. List every operational problem the material reveals — stated or implied. For each, note the evidence (quote or paraphrase from the source material). Discard anything with no evidence; an audit built on assumptions gets torn apart on the client call.
Step 2 — Cost each problem
For every problem, estimate what it costs the business:
- Hours/week lost (admin, follow-ups, quoting, reporting, rework)
- Revenue leaked (missed follow-ups, slow quotes, churn)
- Risk carried (single point of failure, compliance, key-person dependency)
Use the client's own numbers where the conversations contain them. Where you estimate, say so and show the assumption. A defensible rough number beats a precise invented one.
Step 3 — Rank by ROI
Score each problem: ROI = (cost of problem × fixability) / effort to fix. Sort descending. The top 3 are the audit's spine. Everything else goes in an appendix — never bury the lead under twelve findings.
Step 4 — Argue with the operator
Before writing, compare your ranking against the operator's raw read. Where they disagree, say so explicitly and defend your position — or concede and re-rank. This step is the difference between an audit and consultant slop. Surface the disagreement to the user; let them make the final call.
Step 5 — Write the audit
Structure:
- Executive summary — three sentences. The single biggest problem, what it costs, what fixing it is worth.
- Findings, ranked by ROI — for each: the problem, the evidence, the cost, the fix, the effort. One page max per finding.
- Roadmap — the fixes sequenced over 30/60/90 days. Quick wins first.
- Appendix — lower-ROI findings, one line each.
Voice: direct, specific, no hedging, no consulting jargon. Every claim traceable to the source material or a stated assumption.
Step 6 — Emit the goals file
Alongside the audit, write a GOALS.md: the roadmap converted into long-horizon goals with acceptance criteria, ordered by the ROI ranking. This is the file a cheaper model executes against in future sessions — each goal must be self-contained enough to work on without re-reading the audit.
Outputs
audit-<client-name>.md — the client-facing auditGOALS.md — the execution roadmap for follow-on agent sessions
From the Turi Labs AI Agent Learn series — turilabs.in
