Startup Feedback Engine

Turn raw customer feedback into evidence-backed product priorities and credible customer proof.

Startup Feedback Engine is an English-language Codex skill for analyzing support tickets, interviews, surveys, reviews, churn notes, sales calls, community posts, and mixed feedback exports. It keeps every recommendation connected to the exact evidence behind it, then creates a polished interactive Startup Customer Evidence Map.

What it does

• Builds a traceable evidence ledger from messy qualitative feedback.
• Separates pains, jobs, requests, objections, outcomes, praise, churn reasons, and purchase triggers.
• Groups recurring signals without merging or inventing customer quotes.
• Scores product opportunities by frequency, intensity, breadth, recency, strategic fit, commercial relevance, and evidence quality.
• Produces an evidence-backed Now / Next / Later roadmap or a validation sequence.
• Extracts supported claims, customer language, objections, and case-study candidates.
• Treats testimonial permission as a hard gate.
• Shows unsupported claims and tells you what evidence to collect next.
• Generates an interactive, searchable HTML report plus a CSV evidence ledger.

Installation

npx --yes startup-feedback-engine@latest

Restart Codex after installation so it can discover the new skill.

Usage

Attach or point Codex to your feedback files, then use $analyze-startup-feedback.

Use $analyze-startup-feedback in full mode.

Analyze every feedback file in this folder for Acme. I need to decide what to build next and which customer outcomes we can credibly use in our website copy. Create the complete Startup Customer Evidence Map and keep all quotes private unless explicit publication permission exists.

For roadmap decisions only:

Use $analyze-startup-feedback in roadmap mode on these support tickets, churn notes, and interview transcripts. Rank the opportunities, show the evidence IDs behind each recommendation, and give me a 30/60/90-day roadmap only if the evidence supports one.

For customer proof only:

Use $analyze-startup-feedback in proof mode. Find supported outcomes, objections, customer vocabulary, and case-study candidates. Do not turn praise into claims and do not treat permission-unknown quotes as approved testimonials.

How it works

1. Evidence intake

The skill inspects the supplied feedback first. It normalizes source, date, segment, account context, language, and consent only when that metadata exists. Missing information stays unknown.

2. Immutable evidence ledger

Each feedback item receives a stable evidence ID. The exact original wording is preserved separately from translations, paraphrases, and interpretations.

3. Signal classification and clustering

Feedback is classified as pain, job, feature request, objection, outcome, praise, churn reason, purchase trigger, workaround, or counter-evidence. Related items can form a cluster, but every cluster remains traceable to its evidence IDs.

4. Dataset health

Before recommending anything, the skill reports source coverage, date range, segment gaps, duplicates, missing metadata, and likely selection bias. Qualitative feedback is never presented as statistically representative.

5. Product prioritization

Customer problems become opportunity hypotheses rather than automatic feature orders. The skill scores the quality and relevance of the evidence, exposes incomplete scoring, and distinguishes quick wins, research bets, strategic investments, and signals to ignore.

6. Customer proof

Outcomes and customer language are scored separately from product opportunities. Every proof item shows its supporting evidence, qualification, permission state, recommended placement, and what is still needed before publication.

7. Startup Customer Evidence Map

The final standalone report includes:

• Executive verdict
• Dataset health and source coverage
• Customer signal clusters
• Product opportunity matrix
• Evidence-backed roadmap
• Customer proof library
• Objections and case-study candidates
• Unsupported claims and evidence gaps
• Next research questions
• Searchable and filterable evidence ledger

Modes

| Mode | Best for | | --- | --- | | full | Complete evidence map, roadmap, proof library, and report | | roadmap | Product priorities, friction, churn risks, and validation | | proof | Supported outcomes, customer language, objections, and case studies | | churn | Cancellation reasons, failed expectations, and recovery actions | | interviews | Discovery or sales transcripts without treating stated intent as behavior | | compare | Segments, plans, cohorts, time periods, sources, or product versions |

Outputs

outputs/<startup>-feedback-analysis.json
outputs/<startup>-customer-evidence.html
outputs/<startup>-evidence-ledger.csv
outputs/<startup>-roadmap.md
outputs/<startup>-proof-library.md

The HTML report is standalone and can be opened locally. No external JavaScript, analytics, or server is required.

Privacy and quote integrity

• Quotes are never invented, merged, polished, or completed.
• Original non-English text is preserved beside any labeled translation.
• Private identifiers are redacted from shareable reports by default.
• permission-unknown and internal-only quotes are never presented as public testimonials.
• Attractive claims with weak evidence are labeled unsupported.
• Every recommendation, claim, and cluster links back to stable evidence IDs.

Local demo

Generate the included fictional OrbitDesk report and ledger:

npm run demo

Then open outputs/orbitdesk-customer-evidence.html.

Alternative installation

Install directly from GitHub:

npx --yes skills add Kappaemme-git/startup-feedback-engine -g -y --skill analyze-startup-feedback --agent codex

Or copy the analyze-startup-feedback directory into your compatible agent skills directory.

Requirements

• Codex or another agent that supports the Agent Skills format
• Node.js 18 or newer for the report and CSV generators

License

MIT