AI coding assistants are becoming very good at reading code.
But they still struggle with something humans rely on constantly:
understanding why the code exists.
Engineering decisions rarely live in the code itself.
They live in pull requests, review comments, issues, RFCs, and team discussions.
Over time, that context disappears.
Months later, an AI revisits the codebase, sees an unusual implementation, labels it as technical debt, and confidently removes it—without knowing the reason it was introduced.
The consequences can be expensive:
❌ Reintroduced bugs
❌ Broken workarounds
❌ Repeated engineering mistakes
❌ Lost organizational knowledge
To explore this problem, we built GhostPR.
GhostPR is a decision-memory layer for agentic IDEs.
It captures engineering decisions, links them to code, and retrieves historical context before AI-generated changes are applied.
To make memory useful—not permanent—each decision stores:
• Decision
• Reasoning
• Evidence source (PR / Issue / Review / RFC)
• Confidence score
• Current status (Active / Questionable / Deprecated)
Example:
Developer:
Refactor auth/session.ts
GhostPR:
⚠ Historical Context
Decision: Custom OAuth refresh flow
Reason: Provider incompatibility with standard refresh
Evidence: PR #184 discussion
Outcome: Authentication failures reduced after deployment
Status: Active
Recommendation: Review before modifying token expiry logic.
The goal isn’t to tell developers what to do.
It’s to make sure both humans and AI can access the reasoning behind critical engineering decisions.
Because understanding why code exists is often more valuable than understanding what it does.
We’re continuing to improve this and I’d genuinely love feedback, critiques, ideas, or contributions from people thinking about agent memory, developer tools, and AI workflows.
If this resonates with you, let’s build better engineering memory together.
hashtag#AI hashtag#AgenticAI hashtag#DeveloperTools hashtag#SoftwareEngineering hashtag#MCP hashtag#AgentMemory hashtag#OpenSource hashtag#GhostPR
Github Repo: https://github.com/Gaurang774/GhostPR
Top comments (0)