StrayMark adds FollowUps: deferred work now counts

Every project carries two lists. The list of what got done is usually well kept: commits, pull requests, records. The list of what was left for later fares worse — it lives in margin notes, in TODO comments, in the memory of whoever was there that day. StrayMark just gave that second list the same treatment as the first.

What changes

From the start, StrayMark asks every AI-assisted change to leave a record of what was deferred. That much was already being written down; what you couldn’t do was query it with confidence. Answering “what’s pending in this project?” meant digging through files one by one, and the homegrown tools that tried to summarize it lied without knowing it: during the migration, one adopter found twenty-nine real pending items where their old script reported everything was in order.

FollowUps turn that convention into a first-class artifact: each pending item is recorded with its severity and origin, listed and filtered from the command line, and the counters recompute themselves on every write. Deferring something stops being a note and becomes a visible technical decision — which is what it always was.

The third piece of the same idea

In retrospect, the pattern is consistent. First came Charters: declaring scope, verification and risks before executing, and reconciling any deviation before closing. Then came external auditors: several different models independently review the same finished work, so that trust doesn’t hang on a single pair of eyes. Now FollowUps close the triangle: what you decide to do gets declared, what got done gets audited, and what gets deferred counts too.

If you work with AI agents and this problem sounds familiar — decisions that get lost, risks nobody wrote down, pending work that only exists in someone’s memory — StrayMark is a command-line tool that structures that collaboration without taking judgment away from the human. It’s free and open source (MIT license). The full story behind FollowUps, with all the technical detail we deliberately left out here, is on its development blog.