Agent Skill Product Traceability

Product Traceability

Always-active skill that keeps Product Requirements Documents (PRDs), changelogs, decision records, and traceability matrices continuously in sync with every AI-assisted change made in a project.

Install this skill once and every Claude Code session in every project automatically produces an auditable, versioned record of what was requested, what was built, and why.

Overview

Product Traceability ensures that every AI-assisted development activity is accurately documented and linked to evolving product requirements. It automatically maintains PRDs, changelogs, and decision records across agents, models, tools, and sessions.

This skill provides a persistent, auditable source of truth for AI-native product development.

Purpose

To continuously track what was requested, what was built, and what changed — ensuring alignment between intent, implementation, and documentation.

Core Capabilities

• Automatically updates Product Requirements Documents (requirements.md)
• Generates structured and versioned changelogs (change.log)
• Tracks prompts, outputs, and implementation decisions (decisions.md)
• Maintains continuity across AI agents, models, and sessions
• Links requirements to code, commits, and artifacts

(traceability-matrix.md)

• Produces decision logs and rationale records
• Synchronizes documentation across platforms
• Creates an auditable history of AI-assisted work

Files Maintained in Every Project

| File | Role | Update cadence | |------|------|----------------| | requirements.md | Living snapshot of the product (PRD) | Edit in place when behavior changes | | change.log | Append-only, dated history of changes | Prepend a new entry per change | | decisions.md | Architecture / product decisions + rationale | Append when a non-obvious decision is made | | traceability-matrix.md | Requirement → prompt → artifact mapping | Update when features ship or move |

If any of these files do not yet exist in the project root, create them from the templates bundled with this skill.

When to Update

Update the relevant files at the end of every task that modifies the project's code, configuration, or structure. Do not skip this step. If only change.log needs a new entry but requirements.md is already accurate, still verify requirements.md and note that no update was needed.

A dedicated Stop hook also reminds Claude to finalize these files before returning control to the user — see the installer for details.

File Formats

requirements.md — Living PRD

The single source of truth for what the application does right now.

# Product Requirements

## Overview
One-paragraph summary of what the application is and does.

## Features
- **Feature name**: Description of user-facing behavior. _(Status: Implemented | Planned | Deprecated)_
  - Source: session YYYY-MM-DD · prompt excerpt
  - Artifacts: path/to/file.ts, path/to/other.py

## Configuration
- `ENV_VAR` — default, purpose
- Settings and options

## Dependencies
- External services, libraries, runtimes

## Architecture
High-level description of how components are organized and interact.

## Constraints
Known limitations, invariants, requirements.

When a feature is added, add it. When a feature is removed, remove it. When behavior changes, update the description. This file should always answer: "What does this application do right now?"

change.log — Append-Only History

## [Unreleased]

### Added
- Implemented real-time notifications using WebSockets.

### Updated
- Revised notification settings in the PRD.

### Removed
- Dropped legacy polling endpoint.

## 2026-04-21

- Feature X added: short description and why.
- Config Y changed: short description and why.

Newest entries at the top. Group same-day entries under one date heading. When cutting a release, promote [Unreleased] to a dated version heading (Keep a Changelog / SemVer style).

decisions.md — Decision Records

## 2026-04-21 — Chose WebSockets over SSE for notifications

**Context:** Real-time delivery needed for in-app notifications.
**Decision:** Use WebSockets via `socket.io`.
**Rationale:** Bidirectional channel already required for presence.
**Consequences:** New dependency on Redis for pub/sub fan-out.
**Related:** requirements.md §Features/Real-Time Notifications

traceability-matrix.md — Requirement ↔ Artifact Map

| Requirement | Prompt excerpt | Implementation | Decision | Changelog | Verification | |-------------|----------------|----------------|----------|-----------|--------------| | Real-Time Notifications | "Add real-time notifications using WebSockets." | notifications_service.ts | 2026-04-21 | 2026-04-21 Added | tests/notifications.test.ts |

Traceability Model

| Artifact | Description | |----------|-------------| | Requirement | Product feature or capability requested | | Prompt | Instruction given to an AI agent | | Implementation | Generated or modified code and assets | | Decision | Rationale behind changes | | Changelog Entry | Summary of updates | | Session | Context of interactions and outputs | | Verification | Tests, evaluations, or approvals |

Inputs

• User prompts and instructions
• AI-generated code and artifacts
• Agent outputs and session logs
• Existing PRDs and specifications
• Git commits and pull requests
• Project management tickets and tasks

Outputs

• Updated PRDs and specifications
• Structured changelogs
• Decision and audit logs
• Traceability matrices
• Versioned Markdown or JSON documentation
• Session summaries linked to deliverables

Integrations

• Version Control: GitHub, GitLab, Bitbucket
• Documentation: Notion, Confluence, Google Docs
• Project Management: Jira, Linear, Asana, Trello
• AI Platforms: OpenAI, Anthropic, Google Gemini
• Agent Frameworks: MCP-compatible systems
• Developer Tools: Codex, Claude Code, Cursor, Windsurf

Example Commands

• "Update the PRD based on this session."
• "Generate a changelog from today's work."
• "Document what was requested and implemented."
• "Link these changes to their originating requirements."
• "Summarize updates across all agents and models."
• "Create a traceability matrix for this feature."
• "Sync documentation with the latest code changes."

Success Criteria

• Documentation remains synchronized with implementation
• Every change is traceable to its originating request
• Context persists across sessions, models, and tools
• Requirements, code, and decisions stay aligned
• No manual documentation is required
• Outputs are structured, versioned, and auditable

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: vmysla
• Source: vmysla/agent-skill-product-traceability
• License: MIT

Install and usage instructions live in the source repository linked above.