Agent Context

Coming Back to a Project After a Week Away (Without the Dig)

After a break, every project looks unfamiliar and you start by digging through commits and chats. How to leave a project so coming back takes five minutes.

By Tu Nguyen · · 2 min read.md

A developer sits on the living room floor with papers and a laptop, trying to remember where a project stood.

You’re back after a long weekend — or a week, or a month. You open the project. The code looks like someone else wrote it. There are uncommitted changes you don’t remember. The last chat ends mid-thought.

So you dig: commit log, diffs, three old chats, a notebook page. An hour later you’re roughly where you were.

Part 4 of No Process — about leaving projects so you can come back to them.

The break isn’t the problem

Breaks are good. The problem is that most projects are left in a state only your short-term memory understood — and that memory is exactly what the break erased.

The cost isn’t the forgotten work. It’s the re-entry: the time from opening the project to doing something useful.

Leave a re-entry note

Before any break longer than a day, spend five minutes on the state file:

Phase:    6/9 — edit pass
Next:     apply the edit notes to chapter 4 (notes in edits/ch4.md)
Half-done: chapter 3 tables — reformatted 2 of 5
Open:     keep or cut the appendix? (leaning cut)
Session:  resume 3a9…c07 for the edit pass

Then commit and push. This is just writing the next step, plus the two things breaks destroy most: half-done work and open questions.

Coming back: read before you code

When you return, don’t open the code first. In order:

  1. Read the state file. One minute.
  2. Ask the agent to summarize what changed since the date of the last entry — commits, uncommitted files.
  3. Answer the open questions now, while nothing else is loaded.
  4. Start the next step.

The agent does the digging; you do the deciding. It’s the same handoff that stops you re-explaining your project to the AI, just across a longer gap.

A project with no notes

For the old project with nothing written down, don’t start by coding. Make the first session a reconstruction:

“Read this repo and its git history. Draft a state file: what it is, which phase it seems to be in, what looks half-done, and what the next step probably is.”

Correct the draft, commit it, and code in the next session. From then on, the project has memory in files instead of in you.

Where ShotMatic fits

When you open ShotMatic after a break, every project is still there as one line — phase, next step, stuck or not — read from the files in your repos. ✦ resumes the session you left, or starts a new one with the task’s context already written.

FAQ

How do I get back into a coding project after time off?

Start with the written state: phase, next step, open questions. Then ask the agent to summarize commits since that date. Only then open the code.

What should I do before leaving a project for a while?

Spend five minutes updating the state file: where it is, the next step, anything half-done, and which session to resume. Commit and push.

My old project has no notes at all. Where do I start?

Make the first session a reconstruction: ask the agent to read the repo and history and draft a state file, then correct it. Code in the next session.