Reverse

reverse is a Codex skill for reverse-engineering software into product-level specifications.

It is designed for cases where you want to study an OSS project, internal system, API, or product artifact in terms of:

• operator value
• user value
• system capabilities
• observable behavior
• constraints
• non-goals

The goal is to describe what the software does and why it matters without collapsing into implementation details.

Files

• SKILL.md: The actual skill instructions and workflow.
• agents/openai.yaml: UI metadata for the skill name, description, and default prompt.
• references/prompts.md: Reusable prompt pack for different reverse-engineering stages.
• references/rubric.md: Rubric for spotting implementation leakage and rewriting drafts at the product layer.

How To Use

Invoke the skill explicitly in a Codex prompt:

Use $reverse to analyze this repo as a product, not an implementation.
Produce:
1. problem statement
2. actors and goals
3. operator value
4. core capabilities
5. observable behaviors
6. edge cases
7. implementation leakage

You can also ask for narrower outputs:

Use $reverse to extract the operator value and observable behaviors of this system.

Use $reverse to convert these notes into a product spec and flag implementation leakage.

Recommended Workflow

1. Start with product framing.
2. Extract actors and jobs-to-be-done.
3. Group capabilities by outcome.
4. Describe only black-box behavior.
5. Convert observations into testable requirements.
6. Run a leakage review and move technical details out of the product spec.

Prompt Sequence

The usual sequence is:

1. Product Lens
2. Capability Extraction
3. Observable Behavior
4. Requirement Conversion
5. Leakage Review

These prompts live in references/prompts.md.

What Good Output Looks Like

A good reverse-engineered product spec:

• survives a rewrite in another stack
• can be verified by black-box testing
• explains user or operator value
• avoids naming libraries, frameworks, classes, or internal workers unless they are part of the external contract

Reversed Specs

• Claude Subconscious — Persistent background agent that gives Claude Code long-term memory via Letta. Observes sessions, reads codebases, whispers guidance back. (2026-03-28)

Notes

The skill root is ~/src/reverse, not a nested directory. Treat this folder itself as the skill.