Learning from Buffaly sessions

What to capture

A session is a working record: it shows how Buffaly investigated, selected tools, gathered evidence, made decisions, and validated the result. Review it for repeatable patterns, not for every raw detail.

• Capture reusable procedures as workflow memory or prompt skills when they will be repeated.
• Capture stable user preferences or environment facts as ontology memory, not as loose transcript notes.
• Capture documentation gaps when a session reveals that users would need clearer instructions.
• Capture repeated questions, missing validation steps, unclear article names, navigation gaps, and tool choices that worked well or caused confusion.

Turn lessons into improvements

When a session reveals useful knowledge, put it in the place where future work can use it without preserving private or temporary noise.

What not to capture

• Do not store secrets, tokens, passwords, or one-time credentials as memory.
• Do not promote a failed hypothesis into durable guidance without validation.
• Do not turn session-specific scratch notes into global rules unless they apply beyond that task.

End-of-session check

Before asking Buffaly to remember something, separate reusable knowledge from one-time task evidence.

• Keep task-specific files, commit hashes, and validation output in the task or session artifacts.
• Promote repeatable steps into workflow memory only after the approach has worked at least once.
• Store stable environment facts only when they will help future routing or validation.
• Link the remembered lesson back to the source task, wiki article, or public doc when possible.

Review loop

• At the end of a long task, ask what should be remembered, documented, or converted into a reusable skill.
• Prefer small, explicit memories with clear scope over broad behavioral instructions.
• Verify that remembered guidance can be found again through normal Buffaly discovery.

Good review questions