~/.phren/               # or wherever your phren repo lives
  global/               # skills and config for all projects
    skills/             # global skill files (.md)
    CLAUDE.md           # global Claude instructions
  <project>/            # per-project config
    CLAUDE.md
    skills/
    summary.md
  profiles/             # machine profile definitions
    work.yaml
    personal.yaml
  machines.yaml         # maps machine names to profiles

If any of these are missing, the skill will tell you what to create and how.

How it works

Phren uses profiles to decide what goes where. A profile is a list of projects. A machine is mapped to a profile. When you sync, phren only brings over the projects in your machine's profile.

Sync down (pull to this machine)

When the user says "get my skills", "sync my config", "pull from phren", or just "phren-sync":

1. Find the phren directory

PHREN_DIR="${PHREN_DIR:-$HOME/.phren}"
ls "$PHREN_DIR" 2>/dev/null

If the directory doesn't exist, tell the user:

"No phren repo found at ~/.phren. Clone yours with git clone <your-repo-url> ~/.phren, or set PHREN_DIR to point to a different location."

2. Figure out which machine this is

cat ~/.phren/.machine-id 2>/dev/null || hostname

If ~/.phren/.machine-id doesn't exist, ask the user to pick a name:

"I need a machine name to look up your profile. What should I call this machine? (e.g. 'work-laptop', 'home-desktop')"

Then save it:

mkdir -p ~/.phren && echo "work-laptop" > ~/.phren/.machine-id

3. Look up the profile

cat "$PHREN_DIR/machines.yaml"

If the machine name isn't in machines.yaml, tell the user:

"Machine 'work-laptop' isn't in machines.yaml yet. Which profile should it use?" Then show available profiles: ls "$PHREN_DIR/profiles/"*.yaml

cat "$PHREN_DIR/profiles/<profile-name>.yaml"

The profile YAML has:

name: work
description: Work laptop setup
projects:
  - myapp
  - api-server

If no profiles exist yet, offer to create one with phren-init.

4. Pull latest

cd "$PHREN_DIR" && git pull

If this fails (not a git repo, no remote), tell the user. Don't silently skip.

5. Refresh the machine with the supported flow

Run init against the pulled phren repo so hooks, MCP registration, and machine/profile wiring are refreshed:

PHREN_PATH="$PHREN_DIR" phren init -y

If the user is in an untracked repo afterward, tell them to open a session there and let the agent ask, or run phren add from that directory.

Only mention phren link if the user is explicitly repairing an older pre-init install.

6. Build the context file

Write ~/.phren-context.md with a summary of what's active on this machine:

# Phren context: <machine-name> (<profile-name>)

Machine: <machine-name>
Profile: <profile-name>
Last synced: <date>

## Active projects
- myapp: Description from summary.md
- api-server: Description from summary.md

## Global skills
phren-sync, phren-init, phren-consolidate, phren-discover

Pull project descriptions from each project's summary.md if it exists.

7. Rebuild MEMORY.md

Regenerate ~/.claude/projects/*/memory/MEMORY.md with cross-project awareness from the active profile. Pull key facts from each project's summary.md.

8. Check project health

After syncing, check each active project in the phren directory for missing files. Print a one-line status per project. Informational only, does not block the sync.

Required files (warn if missing):

summary.md: project description used by context file and memory
CLAUDE.md: project-level Claude instructions

Recommended files (note if missing, don't warn):

tasks.md: persistent task queue
FINDINGS.md: captured session findings

For each project in the profile, check $PHREN_DIR/<project>/ for these files and print:

Project health:
  + my-app: all files present
  ~ backend: missing recommended: FINDINGS.md
  ! new-project: missing required: summary.md, CLAUDE.md

Legend: + all good, ~ missing recommended files, ! missing required files.

Skip the "global" entry since it has a different structure. Only check actual project directories.

9. Report

phren-sync down: <machine-name> (<profile-name>)

Synced:
  - myapp (CLAUDE.md + 3 skills)
  - api-server (CLAUDE.md + 1 skill)
  - global (CLAUDE.md + 11 skills)

Skipped (not in profile):
  - old-project

Project health:
  + myapp: all files present
  ~ api-server: missing recommended: FINDINGS.md

Context: ~/.phren-context.md updated

Sync up (push changes back)

When the user says "sync this back", "push to phren", "save this to my phren":

1. Find what changed

diff "$PHREN_DIR/<project>/CLAUDE.md" ./<project-path>/CLAUDE.md
diff -r "$PHREN_DIR/<project>/skills/" ./<project-path>/.claude/skills/

Remember: `.claude/skills/` is a generated mirror. The source-of-truth lives in phren `skills/`, and the project mirror also includes inherited global skills after resolution.

If files are symlinked (same inode), they're already in sync. Just commit from the phren directory.

If files are regular copies that differ, copy them back:

cp <changed-file> "$PHREN_DIR/<project>/<corresponding-path>"

2. Commit and push

cd "$PHREN_DIR"
git add -A
git commit -m "<project> updates from <machine-name>"
git push

Conversational shortcuts

These should work naturally in conversation:

User says	What to do
"add myapp to my work profile"	Edit the profile YAML, add to projects list, commit
"show me my personal profile"	Read and display the profile YAML
"what's on this machine?"	Read and display `~/.phren-context.md`
"remove old-project from work"	Edit the profile YAML, commit
"switch to personal profile"	Update the machine mapping in `machines.yaml`, re-run sync down
"create a new profile called travel"	Create a new profile YAML with empty projects list, ask what to include

New machine setup

First time on a new machine:

# 1. Clone your phren repo
git clone <your-repo-url> ~/.phren

# 2. Name this machine
mkdir -p ~/.phren && echo "my-machine-name" > ~/.phren/.machine-id

# 3. Map it to a profile in machines.yaml
#    Add a line: my-machine-name: work

# 4. Run phren-sync

If you don't have a phren repo yet, start with phren-init to create one from scratch.

Conflict resolution

When two machines edit the same phren file before syncing, git pull will hit a merge conflict. Here's how to handle each file type:

tasks.md: Take both changes. Task items from both machines are valid. Concat them, deduplicate if needed, commit.

FINDINGS.md: Take both changes. Keep entries in chronological order. If both machines added entries on the same date, interleave or group them under the same date heading.

CLAUDE.md: Manual merge. Ask the user which version they want, or show both and let them pick. These files contain preferences and instructions where intent matters, so don't auto-resolve.

skills/ files: If the same skill was edited on both machines, show the user a diff and let them choose. If different skills were edited, git handles this automatically (no conflict).

General approach: Run git pull --rebase during sync. If conflicts appear, tell the user which files conflict and apply the rules above. Don't silently drop changes from either side.

Related skills

phren-init: create a new project or set up phren from scratch
add_finding(): capture findings via MCP (synced across machines by push_changes())
phren-consolidate: synthesize findings across projects