eg-prd

name	eg-prd
description	Build a PRD with Codex using codebase grounding, gap-filling, research goldfish, and synthesis.

EG PRD Skill

Build a Product Requirements Document from a rough idea. Use the user's text around the $eg-prd skill mention as the seed. If no seed is provided, ask for one.

If the seed is a GitHub issue URL or #<number>, fetch it with gh issue view <number> and use its title and body as the seed.

Frame

Use Codex structured user input when available. Otherwise ask concise chat questions.

Ask:

1. Depth: lightweight, standard, or comprehensive.
2. Research scope: none, web search only, or web plus Browser Use / in-app browser for gated sources.
3. Output target: save to a project-appropriate PRD path, print/chat only, hand off to new-feature, save durable nuggets to AGENTS.md.

Codebase Grounding

The user invoked this skill, so fresh subagents for codebase grounding are explicitly authorized.

Spawn 1-2 fresh Codex subagents with fork_context: false; use agent_type: "explorer" when available.

• Goldfish A: existing surfaces near the seed. Report file:line citations, data model, routes/endpoints/screens.
• Goldfish B: architecture and conventions. Read AGENTS.md, CLAUDE.md, docs, manifests, and recent commits. Report stack, auth/scoping, testing tiers, deployment topology, and constraints.

The elephant prints:

CODEBASE BRIEF
- Closest existing surfaces:
- Patterns to mirror:
- Constraints from AGENTS.md / CLAUDE.md / docs:
- Observed gaps:

Gap List

Surface gaps grouped by:

• Who and why
• What and scope
• How well
• When
• Constraints
• Failure modes
• Pricing and go-to-market, if relevant
• Instrumentation

Number gaps globally as G1, G2, etc. Ask which gaps to fill before drafting. Unselected gaps go into Open Questions.

For each selected gap, ask a structured question with 3-5 plausible answers plus "defer" and "other".

Research

If research scope is not "none", spawn 3-5 fresh research goldfish in parallel with fork_context: false.

Possible lenses:

• Market and prior art
• Technical patterns
• UX precedent
• Compliance and risk
• Performance and scale

Each goldfish gets only:

You are a fresh researcher with no prior context. Your lens is <LENS NAME>.

SEED: <PASTE USER SEED>
CODEBASE BRIEF: <PASTE>
ANSWERED GAPS: <PASTE>

Research your lens. Produce 5-10 numbered findings with sources, then 3-5 bullets on what this implies for the PRD.

If Browser Use / in-app browser access to a logged-in or sensitive source is needed, stop and tell the elephant what you want to inspect and why.

End with `lens complete`.

The elephant summarizes research by lens and consolidates sources.

PRD

Draft:

# PRD: <title>

## Executive Summary
## Problem Statement
## Target Users
## Current State
## Proposed Solution
## Scope
## User Stories / Jobs To Be Done
## Functional Requirements
## Non-Functional Requirements
## Success Metrics
## Risks & Mitigations
## Implementation Hints
## Open Questions
## Sources & References
## Out-of-Scope Follow-Ups

Skip sections only when genuinely not applicable. Mark deferred gaps as Open Questions.

Review And Output

Ask whether the PRD is approved, needs specific section revisions, should restart, or should stop.

If approved, apply the selected output targets:

• Save PRD if selected. Infer the path at runtime: prefer an existing PRD/spec directory such as docs/prds/, docs/specs/, prd/, or a path documented in AGENTS.md/CLAUDE.md; otherwise use docs/prds/<slug>-<YYYY-MM-DD>.md when docs/ exists, or notes/prds/<slug>-<YYYY-MM-DD>.md.
• Persist only durable cross-cutting nuggets to AGENTS.md if selected; show the diff before applying.
• Print the exact Use $eg-new-feature to build <summary> handoff prompt if selected.

Final Report

Report:

• PRD title
• Depth and research scope
• Codebase brief summary
• Gaps surfaced, filled, deferred
• Research lenses run
• Output location
• Next action

Stop short of staging or committing. Follow the commit policy documented in AGENTS.md/CLAUDE.md if the user later asks to commit.