project memory

The problem

LLM sessions start with a blank slate. You lose hours re-explaining architecture, repeating past decisions, or watching the agent suggest a library you already rejected three weeks ago.

What it does

• Maintains a persistent Markdown graph of project threads, decisions, and outcomes.
• Tracks rejected approaches and the specific conditions needed to revisit them.
• Reads only active and relevant history at session start to preserve context window.
• Automatically creates and manages a .memoria/ directory within your project.
• Flags "stale" decisions as project conditions evolve, acting as a technical anchor.

Why this beats prompting it yourself

Manually pasting context is error-prone and consumes your token budget. This skill implements a structured state machine for project memory, ensuring the agent "remembers" not just what you did, but why you did it, without you managing the files.

Use cases

• Resuming complex coding tasks after days or weeks away from the codebase.
• Preventing agents from suggesting "fix" attempts that have already failed.
• Onboarding a new LLM agent to a long-running project with deep technical debt.
• Tracking architectural trade-offs and "revisit if" triggers for infrastructure shifts.

Known limitations

Requires write access to the local file system. Not intended for one-off tasks or quick questions that finish in a single session.