Skip to content
Back to skills

Requirements Gap Investigator

ASecurity

Investigate a vague or incomplete feature request to surface unstated assumptions, missing constraints, and conflicting stakeholder signals before requirements are written. Use when a feature ask is ambiguous, underspecified, or conflicting; do not use to write a PRD, make product decisions, or evaluate an already-detailed spec.

  • 19 stars
  • 0 votes
  • 0 copies
  • 5 views
  • Added September 19, 2026
ai-agentsgo

Security analysis

A100/100

Pro scans all 3 files and shows the line behind each finding

Scanned September 19, 2026

npx -y skills add khadir-syed/k_ai-agent-skills --skill requirements-gap-investigator --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Requirements Gap Investigator?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Requirements Gap Investigator
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/khadir-syed-requirements-gap-investigator/badge)](https://www.skillsdirectory.com/skills/khadir-syed-requirements-gap-investigator)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: requirements-gap-investigator
description: Investigate a vague or incomplete feature request to surface unstated assumptions, missing constraints, and conflicting stakeholder signals before requirements are written. Use when a feature ask is ambiguous, underspecified, or conflicting; do not use to write a PRD, make product decisions, or evaluate an already-detailed spec.
---

# Requirements Gap Investigator

Determine what is actually known, assumed, or missing about a feature request before anyone drafts requirements from it. Produce an evidence-backed gap report and stop before writing a PRD unless the user separately authorises that.

## Operating boundary

- Treat the task as investigation-only by default. Do not write a PRD, user story, spec, or requirements document.
- Do not decide product direction, prioritise the feature, or resolve a stakeholder conflict on the user's behalf — surface it instead.
- Use only the request text, linked docs, and context the user supplies or that is directly available; do not invent stakeholder intent.
- Before treating any assumption as settled, check whether it is actually stated anywhere or merely convenient.

If the user explicitly asks for both the gap investigation and a drafted PRD in the same request, finish and report the gap investigation first, make the transition to drafting visible, and follow the host's approval requirements for the write itself.

## Investigation workflow

### 1. Frame the request

Record:

- the request as stated, in the requester's own words;
- who is asking and, if known, who else has a stake in the outcome;
- the problem the feature is meant to solve, if stated;
- any deadline or dependency mentioned; and
- any existing docs, tickets, or prior discussion referenced.

Ask only for missing information that blocks a useful investigation. Do not convert an unstated assumption into a stated requirement.

### 2. Establish what is actually known

For every important claim about the request, label it as:

- **Stated:** written down in the request or a linked document, in those words.
- **Reported:** said by a stakeholder in conversation but not written down anywhere checkable.
- **Inferred:** a conclusion drawn from precedent, similar features, or context — not stated by anyone.
- **Unknown:** not established by anything available.

Common gaps worth checking explicitly: who the feature is for, what "done" looks like, what is explicitly out of scope, how it should behave for edge cases (empty states, permissions, scale), and what happens to any existing behaviour it replaces.

### 3. Identify conflicts and gaps

Compare what different stated or reported sources say. Note any place where:

- two stakeholders describe the same feature differently;
- the stated goal and the stated scope don't obviously match;
- a constraint (timeline, platform, compliance) is mentioned once but not accounted for elsewhere; or
- success is not defined in any checkable way.

### 4. Compare interpretations

Where the request is genuinely ambiguous, list the plausible interpretations rather than picking one silently. For each interpretation, note what evidence supports it and what a requirements author would need to confirm before building on it.

### 5. Report and stop

Report:

- investigation status: clear enough to draft from, gaps identified, or too ambiguous to proceed;
- the gaps and conflicts found, each labelled stated/reported/inferred/unknown;
- competing interpretations, where they exist;
- what would need to be confirmed, and with whom, before writing requirements; and
- explicitly, that no PRD or spec has been drafted.

Stop after the report. Do not silently begin drafting requirements from the gaps you found.

## Stopping conditions

Stop and explain the boundary when:

- the gaps are identified well enough to answer the request;
- resolving a gap requires a decision only a named stakeholder can make;
- the request touches a scope, budget, or commitment decision beyond what was asked; or
- the available material is too thin to say anything useful beyond "this needs more input."

## Quality check

Before returning the report, confirm that:

- every gap or claim is labelled stated, reported, inferred, or unknown;
- no unstated assumption was treated as a settled requirement;
- conflicting stakeholder signals are presented as conflicts, not silently resolved; and
- the response does not draft requirements, a PRD, or a product decision.

Files in this skill

  • SKILL.md4.4 KB
  • examples/vague-export-request.md2 KB
  • references/gap-report-format.md1.2 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.