Product Requirements (Interactive PRD Skill)
Turns vague feature ideas into a professional PRD through interactive dialogue and a 100-point quality score, only generating at 90+.
Pure prompt-based workflow: reads project context files and writes only to docs/{feature-name}-prd.md; no network calls, no credential handling, no destructive defaults, least-privilege and transparent data flow, with a final confirmation step. Deduction: no declared handling/rollback for existing files or naming conflicts; overwrite behavior is not explicitly stated.
Instructions are self-consistent with a clear stepwise flow, but the skill depends on the AskUserQuestion tool without a stated fallback if it is unavailable; the 90-point threshold is subjective model scoring with uncertain outcomes; no tests or abnormal-input handling exist for this skill. Static cap of 10 applies.
Scenario (requirements gathering and PRD generation) and trigger conditions are clearly declared in the description, and it explicitly instructs 'think in English, respond in Chinese', which suits Chinese users; core function does not depend on overseas-only services. Deduction: no declared non-fit boundaries (e.g., non-software requirements), and triggers may collide with adjacent skills like test-cases.
Well-structured documentation, complete template, ample example dialogue; repository has AGPL-3.0 license and version number. Deduction: the SKILL.md itself has no independent version, changelog, known-limitations disclosure, or explicit maintenance ownership, and the scoring rubric has no calibration basis.
Can produce a structurally complete, directly usable PRD through dialogue with a professional template; real marginal value over writing manually. Deduction: the 100-point quality score is a subjective heuristic — the 90 threshold may cause excessive clarification rounds or premature gating; static review with no execution evidence, capped at 7.
Only the SKILL.md source is auditable; repository CI and test suites cover codeagent-wrapper and harness but nothing touching this skill's key paths, so no third-party execution evidence exists. Static cap of 5 applies; only source-level audit achieved.
- The 90-point quality threshold is a subjective model judgment; the score is not objectively calibrated and should not be treated as a hard gate for requirements completeness.
- Overwrite/rollback behavior when writing docs/ is undefined if a same-named file exists; back up before use.
- The skill depends on the AskUserQuestion tool; behavior without it is unspecified.
- Publisher is not verified by the FollowSkills registry; identity is unknown; only the AGPL-3.0 license is verifiable.
- This is a purely static source review with no execution verification; confidence is low.
What does this skill do, and when should you use it?
product-requirements is one of 11 skills bundled in the myclaude repository, located at skills/product-requirements/. It plays the role of Sarah, a meticulous Product Owner: it first reads your project's README and package./pyproject.toml to build context, then scores your requirements across five dimensions (business value, functional requirements, UX, technical constraints, scope). Below 90 points it asks targeted clarifying questions and iterates; at 90 or above it generates a full PRD and saves it to docs/{feature-name}-prd.md. The entire flow is dialogue-driven — no shell commands, just local file reads and writes.
1) Reads the project README and package./pyproject.toml in parallel to understand the tech stack and conventions; 2) scores requirements on a 100-point, five-dimension rubric (Business Value 30, Functional 25, UX 20, Technical Constraints 15, Scope 10); 3) when below 90, uses the AskUserQuestion tool with 2-3 targeted questions per round and recalculates iteratively; 4) once at 90+, generates a PRD from a detailed template including executive summary, user stories with acceptance criteria, MVP phasing, and a risk table; 5) saves the document as docs/{feature-name}-prd.md.
- A solo developer who wants to turn a fuzzy feature idea in their head into a structured requirements document before writing code
- A small team without a dedicated PM that needs a fixed scoring rubric to force requirements clarity and keep vague specs out of development
- A project lead who wants a handoff-ready PRD with user stories, acceptance criteria, and an MVP scope before dev work begins
- Someone inheriting an unfamiliar codebase who wants new-feature requirements discussed in the context of the existing stack, via the skill's context-reading step
What are this skill's strengths and limitations?
- The 90-point quality gate systematically prevents vague requirements from being turned straight into a document
- The five-dimension score breakdown is shown transparently, and you can watch the score climb after each answer
- The PRD template is thorough, covering user stories, acceptance criteria, MVP phasing, a risk table, and dependencies
- Implemented purely via dialogue plus file read/write — no external dependencies, works out of the box
- Requires patience for multiple Q&A rounds; users wanting a quick document will be blocked by the threshold
- The SKILL.md explicitly references Claude Code's AskUserQuestion tool, so porting to other clients requires editing that reference
- The scoring rubric is a hardcoded heuristic — a high score doesn't guarantee genuinely good requirements
- The source material provides no test suite or usage data backing the real-world quality of the generated documents
How do you install this skill?
The skill installs with the myclaude collection: run npx github:stellarlinkco/myclaude for the interactive installer, or npx github:stellarlinkco/myclaude --list to see installable items. Alternatively, manually copy the skills/product-requirements/ folder into the Claude Code skills directory (default ~/.claude/skills/). The README documents no standalone install command for this single skill.
How do you use this skill?
Invoke it in Claude Code with a trigger-style request such as 'Help me write a PRD for a dark mode feature' or 'I need to spec out product requirements for this new feature.' It greets you as Sarah, gathers project context, then enters a score-ask-iterate loop; the PRD is only generated at docs/{feature-name}-prd.md once the score reaches 90+. Note the SKILL.md instructs the assistant to respond in Chinese to the user.
How does this skill compare with similar options?
The same repository's agents/requirements/ module offers a 'lightweight requirements-to-code pipeline' (/requirements-pilot command) focused on pushing requirements straight into code; this skill stops at the PRD deliverable, better suited to teams that want a formal requirements document before development starts.