Add domain values layer to CLAUDE.md #1

New Issue

2026-08-07T20:14:25-04:00

jared commented

2026-08-07 20:14:25 -04:00

Wire this repo into the shared values framework so agents working here load the
right domain context automatically instead of relying on judgment.

What to do

Add this section to this repo's CLAUDE.md (create the file if it does not exist):

## Domain context — read this first

**Before doing any work in this repo, read `~/.claude/memory/values/infrastructure.md`.**
It carries the purpose, direction, and constraints for this domain. It is not optional
context — read it before planning or writing code, not after. (An `@` import will not
work here; project-level imports don't preload. The read is the mechanism.)

Layer for this repo: infrastructure

Why an instruction rather than an `@` import

Verified empirically on 2026-08-07: @ imports resolve in ~/.claude/CLAUDE.md but do
not preload from a project-level CLAUDE.md, even in a trusted workspace. An
imperative instruction does work — the agent reads the file when told to.

Prerequisite (already done globally)

~/.claude/memory/values/ is in permissions.additionalDirectories as of dotfiles3
commit b0698b9. Without it the read is permission-blocked and fails silently.

Verify

From a session in this repo:

claude -p "What domain context applies to this repo, and where did you read it?"

The agent should read ~/.claude/memory/values/infrastructure.md and answer from it.

Wire this repo into the shared values framework so agents working here load the right domain context automatically instead of relying on judgment. ## What to do Add this section to this repo's `CLAUDE.md` (create the file if it does not exist): ```markdown ## Domain context — read this first **Before doing any work in this repo, read `~/.claude/memory/values/infrastructure.md`.** It carries the purpose, direction, and constraints for this domain. It is not optional context — read it before planning or writing code, not after. (An `@` import will not work here; project-level imports don't preload. The read is the mechanism.) ``` Layer for this repo: **infrastructure** ## Why an instruction rather than an `@` import Verified empirically on 2026-08-07: `@` imports resolve in `~/.claude/CLAUDE.md` but do **not** preload from a project-level `CLAUDE.md`, even in a trusted workspace. An imperative instruction does work — the agent reads the file when told to. ## Prerequisite (already done globally) `~/.claude/memory/values/` is in `permissions.additionalDirectories` as of dotfiles3 commit `b0698b9`. Without it the read is permission-blocked and fails silently. ## Verify From a session in this repo: ``` claude -p "What domain context applies to this repo, and where did you read it?" ``` The agent should read `~/.claude/memory/values/infrastructure.md` and answer from it.

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: jared/dropbox_helper#1

Add domain values layer to CLAUDE.md #1

What to do

Why an instruction rather than an @ import

Prerequisite (already done globally)

Verify

Why an instruction rather than an `@` import