What is since?

Coding agents like Claude Code, Cursor, and Copilot read a file, reason about it across many steps, then write changes back. But the file may have changed on disk in the meantime: a formatter ran, a teammate pushed code, another agent edited the same file. The agent never notices and overwrites the newer version with full confidence. This is called the stale world model problem, and it accounts for a significant share of agent failures on long tasks. since is a lightweight Python package that solves this. It fingerprints every file the agent reads by recording the file's modification time and a content hash. On every subsequent tool call, it compares those stored fingerprints against the current state of the file on disk. If something changed, since reports it before the agent acts, so the agent re-reads the file instead of writing from a stale copy. There is no background process or polling. It just checks disk at the next turn. The package runs as a local MCP server, which is a standard way for agent tools to communicate. You add it to your client's configuration and add one line to your agent's instructions telling it when to call the provided tools. The main tool is check_staleness, which reports not just the file you asked about but every tracked file that has drifted. This means the agent cannot stay blind to a change it did not think to check. The package has zero dependencies and requires Python 3.10 or later. Beyond file checking, since also offers temporal context for chat applications. You wrap your chat function with a decorator, and the model receives a small time block before each turn showing when the conversation started, how long the gaps were between messages, and whether context has gone stale. This lets the model say "welcome back, it has been about six hours since we last spoke" instead of claiming it has no memory of previous interactions. The package is installed with pip install pysince. The PyPI name is pysince because the name since was already taken on the package index.