okf memory

You lose project context every time you start a new session or switch tools. This skill implements the Open Knowledge Format (OKF v0.1) to maintain durable, agent-readable project memory in a plain-text filesystem.

What it does

• Persistent storage saves decisions, findings, and runbooks to ~/.okf-memory using YAML-frontmatter Markdown.
• Change discovery tracks architectural shifts, modified endpoints, and schema migrations automatically.
• Progressive disclosure organizes knowledge through hierarchical indexes so agents only load relevant notes.
• History logging maintains a chronological record of changes across all projects or within specific ones.
• Cross-session recall ground answers in past project state by reading discovery manifests and dated logs.

How it works

1. The agent identifies the project slug and checks ~/.okf-memory/index.md for existing context.
2. Use the save command to record specific documents like Decision, Schema, or Metric types.
3. Use the discover command to inspect git status and file modifications, updating the discovery.md manifest.
4. Retrieve knowledge by walking YYYY/MM directories or grepping for specific tags and types in the YAML frontmatter.

Frameworks & tools

The skill uses a Python helper script to manage files. It relies on standard filesystem operations, Grep for searching, and Git for change detection.

Why this beats prompting it yourself

Manual prompts forget details as soon as the context window flushes. This skill enforces a standardized directory structure and metadata schema that ensures your agent can query its own history without you re-explaining the codebase every morning.

Use cases

• Record why a specific database was chosen to prevent re-litigating the decision later.
• Track what new changes occurred in a repository between two development sessions.
• Maintain a live runbook of project-specific commands and environment setups.
• Generate a session summary that another agent can consume to pick up where you left off.

Known limitations

Knowledge can become stale if not updated; agents must verify file-based notes against the current codebase. Does not store secrets, tokens, or credentials.