← All guides

Second brain · July 11, 2026 · 3 min read

Let agents keep your goals file current

Goal docs rot because updating them is manual. Make yours derived files: you capture raw thoughts, a scheduled agent recompiles goals and priorities.

The problem

Every goals document I maintained by hand went stale. Life moved, the doc did not, and everything built on top of it inherited the staleness. The failure is structural: keeping a goals file current is manual maintenance, and manual maintenance is exactly the work that quietly stops happening.

It gets worse once AI is involved. If your agents read a priorities file from three months ago, every answer they give is optimised for a life you no longer live.

How do you keep a goals file current?

You stop editing it. In my vault, the profile, goals, and priorities files are derived files: outputs a scheduled agent regenerates, not documents I maintain. The pattern is capture, derive, read.

  1. Capture. I dump raw thoughts from anywhere into one append-only inbox folder, mostly from my phone. A new goal is just a line starting with goal:. Messy is fine.
  2. Derive. The same scheduled agent that maintains my second brain also owns the personal layer. On each run it updates my self-model only when there is real evidence, reconciles the goals page against what the captures show (promoting new goals, marking old ones achieved or dropped, never inventing goals I did not state), and regenerates the ranked priorities list from profile plus goals plus open tasks plus recent captures. It surfaces one starred top item and stamps the file with a last generated date.
  3. Read. Every AI session I open pulls my current state from those derived files. Fresh context, zero upkeep from me.

The one rule that makes it work: never hand-edit a derived file. If the priority order looks wrong, the fix is upstream. Change the goals page or drop a new capture, then let the agent recompute.

Goals are the precious input you set. Priorities are the output the machine computes.

Paste this into the instructions file of any vault agent:

You maintain three derived files: profile.md (how I work),
goals.md (ranked active goals), priorities.md (the to-do list).
On every run:
1. Read the new captures in raw/.
2. Update profile.md only on real evidence. Do not churn it.
3. Reconcile goals.md: promote any capture starting "goal:",
   mark goals achieved or dropped, never invent goals.
4. Regenerate priorities.md from profile + goals + open tasks
   + recent captures, ranked by goal alignment. Surface ONE
   starred top item. Stamp it "Last generated: <date>".
I never hand-edit these files. If I disagree with the output,
I change goals.md or add a capture, and you recompute.

Do this now

  1. Add a header to your existing goals file: “This file is derived. The agent regenerates it. Durable changes go in captures.”
  2. Paste the instruction block above into your vault agent’s instructions.
  3. Run it manually once and check the output. Then put it on a schedule and stop maintaining goal docs forever.

Claude Code

Turn this into a Claude skill

Paste this into Claude Code. It reads this guide and builds a skill shaped to how you work.

Read the guide at https://aiwithamirthan.com/guides/let-agents-keep-your-goals-file-current. Then build me a Claude Code skill from it, personalized to how I actually work.

1. Fetch and read the full guide at the URL above.
2. Look at my setup so the skill fits me, not a generic template: skim my repos, my CLAUDE.md / AGENTS.md files, my existing skills, and my stack.
3. Create a new skill (a SKILL.md with a clear name, description, and steps) that lets me do what this guide teaches, adapted to my tools and goals.
4. Tell me where you saved it and how to invoke it.

If anything about my setup is unclear, ask me one or two questions first.

Want to build this yourself?

Join Claude Code Academy: courses, build alongs, and builders turning AI into real systems. And I post a new build on Instagram nearly every day.