Claude Code at Stratum

Overview

Claude Code is the primary interface for Stratum's operational intelligence. Skills live in plugins, plugins live in marketplaces, and marketplaces are configured per-user in ~/.claude/settings.json. Stratum uses two namespaces:

The shared claude-team-config repo holds: SC-* plugin sources, the workspace CLAUDE.md, shared agents, slash-command templates, and a setup.sh bootstrapper.

First-time setup

Run these once when joining the team. After this, day-to-day updates are just git pull in the team-config repo.

1. Create the workspace and clone the team repo

   mkdir -p ~/REPOS/Stratum\ Collective
   cd ~/REPOS/Stratum\ Collective
   git clone https://github.com/stratum-collective/claude-team-config.git

2. Run the setup script

   bash claude-team-config/setup.sh

   The script handles workspace symlinks, the workspace CLAUDE.md, plugin cache installation, slash-command symlinks, and shared agents. It prints the exact JSON to merge into your settings file in step 3.

3. Merge the marketplace block into ~/.claude/settings.json

   Open ~/.claude/settings.json (create it if missing) and merge in both blocks — do not replace the whole file if it already has content.

   # Shape of what to add (use the script output for real paths):
   {
     "extraKnownMarketplaces": {
       "stratum-team": {
         "source": {
           "source": "directory",
           "path": "/Users/YOUR-USERNAME/REPOS/Stratum Collective/claude-team-config/plugins"
         }
       }
     },
     "enabledPlugins": {
       "SC-operations@stratum-team": true,
       "SC-messaging-quality@stratum-team": true,
       "SC-notion-ops@stratum-team": true
     }
   }

4. Restart Claude Code

   Skills and slash commands are now available. Test with /SC-operations:daily-standup or ask Claude to "run ops-triage".

What the script does automatically

Keeping up to date

When skills are updated (new checks, changed logic, new skills added):

cd ~/REPOS/Stratum\ Collective/claude-team-config
git pull
bash setup.sh    # refreshes cache copies with latest skill files

Plugin lifecycle

Plugin source lives in one place; each user's Claude Code reads from a per-user cache copy. Any source edit must sync to cache or the change won't show up in the running session.

Health check

Run before any rename, install, or major refactor, and after any edit to plugin source:

bash ~/.claude/plugin-health.sh

Checks seven things:

1. Every registry installPath directory exists
2. PL source → cache is current (no newer source files)
3. SC source → cache is current (no newer source files)
4. No legacy @local registry entries
5. Every SKILL.md name: field matches {plugin-dir}:{skill-slug}
6. Every backtick-quoted skill reference in scheduled tasks resolves to a known plugin
7. No orphaned cache dirs in either marketplace

Adding a new plugin

SC plugin (shared):

# 1. Create source dir in team config
mkdir -p ~/REPOS/Stratum\ Collective/claude-team-config/plugins/SC-{name}/skills/{skill-slug}

# 2. Write SKILL.md — name field must be SC-{name}:{skill-slug}

# 3. Sync to cache
mkdir -p ~/.claude/plugins/cache/stratum-team/SC-{name}/unknown
rsync -a ~/REPOS/Stratum\ Collective/claude-team-config/plugins/SC-{name}/ \
  ~/.claude/plugins/cache/stratum-team/SC-{name}/unknown/

# 4. Register in installed_plugins.json
# Add: "SC-{name}@stratum-team" → installPath: cache/stratum-team/SC-{name}/unknown

# 5. Commit so teammates get the new plugin
cd ~/REPOS/Stratum\ Collective/claude-team-config
git add plugins/SC-{name} && git commit

# 6. Health check, then restart Claude Code
bash ~/.claude/plugin-health.sh

PL plugin (personal):

# 1. Create source dir
mkdir -p ~/.claude/plugins/PL-{name}/skills/{skill-slug}

# 2. Write SKILL.md — name field must be PL-{name}:{skill-slug}

# 3. Sync to cache
mkdir -p ~/.claude/plugins/cache/patrick-local/PL-{name}/unknown
rsync -a ~/.claude/plugins/PL-{name}/ ~/.claude/plugins/cache/patrick-local/PL-{name}/unknown/

# 4. Register in installed_plugins.json
# Add: "PL-{name}@patrick-local" → installPath: cache/patrick-local/PL-{name}/unknown

# 5. Health check, then restart Claude Code
bash ~/.claude/plugin-health.sh

Renaming a plugin

Follow exactly — skipping steps causes the cache/registry split that broke the WS-venture-studio rename in 2026-04-24.

# PL example (substitute SC paths and stratum-team marketplace as needed)

# 1. Rename source dir
mv ~/.claude/plugins/OLD-name ~/.claude/plugins/NEW-name

# 2. Create new cache dir and sync
mkdir -p ~/.claude/plugins/cache/patrick-local/NEW-name/unknown
rsync -a ~/.claude/plugins/NEW-name/ ~/.claude/plugins/cache/patrick-local/NEW-name/unknown/

# 3. Delete old cache
rm -rf ~/.claude/plugins/cache/patrick-local/OLD-name

# 4. Update installed_plugins.json key + installPath

# 5. Update SKILL.md name: fields
find ~/.claude/plugins/NEW-name -name SKILL.md -exec \
  sed -i '' 's/OLD-name:/NEW-name:/g' {} \;

# 6. Update internal references in plugin files
find ~/.claude/plugins/NEW-name -type f \( -name "*.md" -o -name "*.json" -o -name "*.sh" \) -exec \
  sed -i '' 's/OLD-name/NEW-name/g' {} \;

# 7. Grep for stale references in scheduled tasks + config
grep -r "OLD-name" ~/.claude/scheduled-tasks/ ~/.claude/CLAUDE.md 2>/dev/null

# 8. Health check must pass before restarting
bash ~/.claude/plugin-health.sh

# 9. Restart Claude Code

Editing a plugin skill

Any source edit creates cache drift. Always sync after editing:

# SC
rsync -a ~/REPOS/Stratum\ Collective/claude-team-config/plugins/SC-{name}/ \
  ~/.claude/plugins/cache/stratum-team/SC-{name}/unknown/

# PL
rsync -a ~/.claude/plugins/PL-{name}/ ~/.claude/plugins/cache/patrick-local/PL-{name}/unknown/

The health check detects drift automatically — run it after edits to see which plugins need syncing.

Deleting a plugin

# 1. Remove source dir
rm -rf ~/.claude/plugins/PL-{name}                                                # PL
rm -rf ~/REPOS/Stratum\ Collective/claude-team-config/plugins/SC-{name}           # SC

# 2. Remove cache
rm -rf ~/.claude/plugins/cache/patrick-local/PL-{name}                            # PL
rm -rf ~/.claude/plugins/cache/stratum-team/SC-{name}                             # SC

# 3. Remove registry entry from installed_plugins.json

# 4. Check for stale references
grep -r "{name}" ~/.claude/scheduled-tasks/

# 5. Health check
bash ~/.claude/plugin-health.sh

Teammate sync (SC plugins)

Teammates pull claude-team-config and run:

cd ~/REPOS/Stratum\ Collective/claude-team-config
git pull
bash setup.sh

setup.sh syncs SC-* sources into the teammate's local cache and registers them in their installed_plugins.json. Any rename committed to team-config requires teammates to re-run setup.sh. PL-* plugins are not synced — they remain personal to Patrick's machine.

What the health check does not cover

• claude-plugins-official — managed by Anthropic, treat as read-only
• Browser/service worker cache — stale HTML after a deploy; force-refresh or unregister the SW manually
• Scheduled task prompt correctness — validates plugin sources exist, not that skill logic is current
• Remote (CCR) triggers — managed at claude.ai/code/scheduled/; health check has no visibility into remote trigger state

Active plugin index

As of April 2026. Three shared SC plugins (20 skills total) plus two personal PL plugins.

SC-operations SHARED

Operational intelligence — standup, triage, milestone gates, reporting, hygiene sweeps. Fifteen skills:

SC-messaging-quality SHARED

Phrase registry and copy-quality tooling. Three skills:

SC-notion-ops SHARED

Notion workspace quality and formatting. Two skills:

PL-venture-studio PERSONAL

Wherewithal venture incubation. Skills: build, craft. Listed for completeness — not shared with Stratum teammates.

overnight-design-scan PERSONAL

Portfolio design quality monitoring across the Wherewithal ventures. Personal, not shared.

stratum-cli reference

The stratum-cli repo (separate from claude-team-config) is the Node.js CLI that handles operational data flow — Notion read/write, lead pipeline, outreach, dashboard data generation. Skills call it; you can also call it directly.

cd ~/REPOS/Stratum\ Collective/stratum-cli
node bin/stratum.js --<command>

--daily-run chains the full pipeline (lead-triage → ... → UIS scan → ops-dashboard-data → pipeline-dashboard-data → push-dashboard-data) and auto-commits the regenerated data.json files to this repo. See Automation Schedule for when each command runs in production.

Conventions & gotchas

Notion API

• content_search_mode: "workspace_search" is required on every notion-search call. The default ai_search floods results with Google Calendar events instead of Notion pages.
• notion-update-page Assignee requires the user://UUID prefix — bare UUIDs fail silently. Patrick: user://2dad872b-594c-817a-9a10-000221456611. Gennovva: user://2d9d872b-594c-8191-bf28-0002c2a3039e.
• Never bulk move/archive Notion pages without explicit page-by-page confirmation. The 2026-03-20 incident trashed the entire workspace during a targeted cleanup.

Page naming

• WF: prefix = workflow procedure (step-by-step execution an agent or human follows).
• OPS: prefix = operational/procedural (execution plans, deployment plans, dated reports).
• No prefix = reference/guide (canonical definitions, system docs, indexes, registries).
• Decision test: "execute steps" → WF:/OPS:. "How the system works" → no prefix.

Page icons

• Every new Notion page must have a page icon (emoji or custom).
• No emojis in page titles. Title is text-only; the icon carries the visual.

Branch & commit

• Imperative tense, concise, descriptive (e.g. "Add denial pattern classifier").
• Commit freely — tell Patrick what changed and why in the next message.
• Never skip hooks (--no-verify) unless explicitly requested.

Troubleshooting

Retired · archive only