Skip to main content

Work order

Stop your coding agents from drifting on long coding sessions

It reads your plan, reads your commits, and tells you whether the work is still the thing you asked for — with the hashes and file lines behind every claim.

Grab the skill
Stop Your Coding Agents From Drifting On Long Coding Sessions
Rev
v3.4.1
Job
Did We Drift
For
Claude Code
Platform
macOS + Windows
Rate
Free — members

Run /did-we-drift and find out. It reads your plan, reads your commits, and tells you whether the work is still the thing you asked for — with the hashes and file lines behind every claim.

Grab the skill

Free for members. It's a zip you hand straight to Claude Code.

Checking your access…
Claude

DOWNLOAD FOR

Claude Code

Skill for Claude Code

⬇ Download Claude Code skill v3.4.1 (.zip)

Unzip it, then paste one line into Claude Code (below).

Install it

Download the zip, unzip it, then paste one line into Claude Code.

1Download the zipLog in with a free member account, tick the disclaimer, and download Did We Drift.
2Unzip itExtract the zip so the folder lands in ~/Downloads/did-we-drift.
3Paste this into Claude CodeHand Claude Code the unzipped folder and this line.Install the Claude Code skill in ~/Downloads/did-we-drift: copy skills/did-we-drift to ~/.claude/skills/did-we-drift. Confirm /did-we-drift appears in my skills.
4You're done whenYou're done when /did-we-drift appears in your skills.

Use it

Init once after you write the plan, then run the check at the end of a session. Open the dashboard for planned vs built; at a milestone, run /did-we-drift update.

Paste into Claude Code
/did-we-drift init /did-we-drift /did-we-drift dashboard
Sign in to keep track of what you've copied

What it does

1

Finds what you declared Limited searches locate your planning documents — even ones git is set to ignore — and those get flagged, because a later session that only has the project files never sees them.

2

Never invents the goal It quotes and cites what you actually said; it will not infer the plan from the code, because that would treat every drift ever committed as the plan.

3

Grades the work and the plan separately The work and the plan are scored separately, so a messy plan is never graded as drift, and one problem is never counted twice.

4

Checks who wrote the plan It uses git history to tell who wrote each goal, so an agent-written plan is not read back to you as your own intent.

5

Runs the done claims Completion claims are sampled and actually run; what cannot be run is reported unverifiable rather than quietly passed.

Good to know

Init is better, not required — skip it and the skill rebuilds a baseline from what it can find, then tells you it is not yet confirmed by you. It will not declare on track when there was nothing to compare against. The dashboard is a view: if the two disagree, the plan wins. The dashboard checker needs Node 20.6 or newer.

Still deciding? These are the usual questions.

Don't sleep on what's next. Free tools, prompts, and skills for Claude Code, Codex, and OpenClaw — sent when they drop. No spam, unsubscribe anytime.

Your agent won't tell you it wandered off. Ask it.

Don't sleep on AI.

Changelog

3.4.12026-08-27Current

v3.4.1 — members-only zip on this page

The skill is now a members-only download from this page. Log in with a free account, tick the disclaimer, and grab the zip.

3.5.02026-08-09

Update keeps the plan itself honest, not just the audit

Old plan files rot: a note that says "fix before stage 10" is still sitting there when stage 10 starts. The update command checks the work first and refuses to refresh over unresolved drift, then adds a new dated entry to the facts and walks every open note and waiting-on-you item to a recorded outcome — resolved, carried, or escalated. Nothing dies quietly in a hand-off.