claude

Manages symlinks between <brain>/claude/ and ~/.claude/, keeping your Claude Code configuration version-controlled in the brain. Also provides a skill manager for discovering and activating Claude Code skills.

Plugin — opt-in

Enable this plugin in your brain config:

plugins:
  settings:
    claude:
      enabled: true

Managed targets

The plugin tracks these items and keeps them symlinked from the brain into ~/.claude/:

Target	Brain path	Link style
CLAUDE.md	brain/claude/CLAUDE.md	single symlink
settings.json	brain/claude/settings.json	single symlink
plans/	brain/claude/plans/	whole directory
memory/	brain/claude/memory/	whole directory
hooks/	brain/claude/hooks/	each child individually
skills/	brain/claude/skills/	each child individually
scripts/	brain/claude/scripts/	each child individually

overseer claude setup

Interactive wizard that adopts existing Claude config into the brain and creates the symlinks. Safe to re-run — already correct links are skipped.

overseer claude setup
overseer claude setup --dry-run   # preview without making changes

The wizard scans each managed target and shows what action it will take:

Action	Meaning
skip	link is already correct
link	brain file exists, local missing — creates symlink
adopt	local file exists (not a symlink) — moves it to brain, then symlinks
migrate	symlink points to old brain path — moves brain file, updates symlink
conflict	both locations have content — manual resolution needed

overseer claude list

List the symlink status of all managed targets without making any changes.

overseer claude list

overseer claude skills

Manage Claude Code skills. Skills are directories containing .md files with a YAML frontmatter description field, placed in <brain>/claude/skills/ or an external search path. Enabling a skill creates a symlink in ~/.claude/skills/ so Claude Code can discover it.

list

Show all available skills with their status.

overseer claude skills list

Status icons:

• ✓ active (symlink in place)
• ○ disabled
• ! in brain but not yet synced — run sync
• · available from an external path, not yet enabled

sync

Sync skill symlinks to match the current enabled/disabled state. Brain skills are linked automatically; external skills are only linked if you have previously enabled them.

overseer claude skills sync

enable / disable

Enable or disable skills by name. Enabling creates the symlink in ~/.claude/skills/; disabling removes it.

overseer claude skills enable my-skill
overseer claude skills enable skill-a skill-b   # multiple at once
overseer claude skills enable                   # interactive multi-select

overseer claude skills disable my-skill
overseer claude skills disable                  # interactive multi-select

list-path

List all registered external skill search paths.

overseer claude skills list-path

add-path

Register a directory as an additional skill search path. If the given directory contains a skills/ subdirectory, that subdirectory is used automatically.

overseer claude skills add-path ~/repos/my-ai-config
overseer claude skills add-path ~/repos/my-ai-config/skills

After adding, the command prints the discovered skills available to enable.

remove-path

Remove a directory from the skill search paths. Also removes any active symlinks in ~/.claude/skills/ that pointed into that directory.

overseer claude skills remove-path ~/repos/my-ai-config/skills
overseer claude skills remove-path   # interactive select

overseer claude mcp

Manage the Claude Code MCP server registration for overseer. These commands delegate to claude mcp add/remove (user scope, available in all projects) and write to ~/.claude.json.

overseer claude mcp install    # register — idempotent
overseer claude mcp uninstall  # remove   — idempotent

Running install when already registered, or uninstall when not registered, prints an informative message and exits cleanly.

The install command writes the full binary path so Claude Code can find it regardless of the PATH environment it inherits.

overseer claude worklog

Capture what happened in each Claude Code session, so the day can be reconstructed later without re-reading transcripts. A single day routinely produces millions of tokens of transcript across several sessions — far too much to summarise in one pass.

overseer claude worklog install     # register the SessionEnd hook
overseer claude worklog show        # today's captured sessions
overseer claude worklog show --date 2026-08-18 --format json
overseer claude worklog uninstall   # remove the hook

install writes <brain>/claude/hooks/session-end-worklog.sh, symlinks it into ~/.claude/hooks/, and registers a SessionEnd hook in <brain>/claude/settings.json. Only the hooks key is rewritten — every other setting keeps its original order and formatting. Both commands are idempotent, and uninstall restores the file exactly.

What a record holds

One record per session per date — a session that resumes the next day produces two. Records are written to ~/.local/share/overseer/worklog/YYYY-MM-DD.jsonl, which is machine-local and deliberately not brain-synced: the brain carries the hook, the machine keeps the data.

Field	Notes
started_at / ended_at	From record timestamps, never file mtime — sessions resume across days
workdirs	Every distinct cwd resolved to its git root and deduped; one session often spans several repos
prompts	Human prompts only — gated on promptSource, minus IDE notifications, slash-command echoes and hook output
files	From Claude Code's file history. Partial by design: edits made through Bash never appear, so treat git as authoritative

Commits are deliberately not recorded here. Work frequently lands in repos that were never a session cwd, so commits are gathered day-wide elsewhere rather than attributed per session.

The hook never fails a session: a malformed payload or unreadable transcript prints to stderr and exits 0.

Config

External skill search paths are stored in the brain config under integrations.claude.skill_search_paths:

integrations:
  claude:
    skill_search_paths:
      - ~/repos/my-ai-config/skills

Disabled skill names are persisted in <brain>/claude/skills-state.json (managed automatically — do not edit by hand).