Skip to main content
skill CX Codex

ecosystem — Claude Code skill: what it does, how to install

ecosystem audits your Claude Code environment by checking skill staleness, project git activity, CLAUDE.md drift, and Mac Mini services.

MIT 115 installs ★ 414 updated 2026-10-23

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add glebis/claude-skills --skill ecosystem -a claude-code
Open on GitHub

What ecosystem does

The skill runs four audit checks directly in bash and gathers the output into an attention rollup. Step one inspects ~/.claude/skills/ to identify broken symlinks and categorize installed skills into active (under 30 days), recent (30 to 90 days), or stale based on file timestamps.

Next, the audit parses ~/ai_projects/ to inspect git activity across all repositories. It flags missing CLAUDE.md files and measures instruction drift by comparing the last commit date of CLAUDE.md against the latest commit in the repository, warning if the guide lags by more than 90 days or remains untracked.

The final step contacts mac-mini over SSH to ping ports 8080 and 8081, tally active third-party LaunchAgents, and read the health.db size. Without the hardcoded ~/ai_projects/ directory structure and a configured mac-mini SSH host, the project checks yield empty results and the hardware step reports UNREACHABLE.

What it helps with

  • Detect broken symlinks and stale folders in ~/.claude/skills/
  • Highlight git repositories lacking a CLAUDE.md instruction file
  • Flag CLAUDE.md files that lag behind project commits by 90 days
  • Check local HTTP ports and custom LaunchAgents on a Mac Mini

Example prompts for the agent

  • /ecosystem
  • Audit my ecosystem health
  • What skills and projects are stale in my setup?

How to install ecosystem

1Skills CLI (Vercel)

npx skills add glebis/claude-skills --skill ecosystem -a claude-code

2Manually: put the files in place

~/.claude/skills/ecosystem/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/ecosystem/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about ecosystem

How to install ecosystem in Claude Code?

Run npx skills add glebis/claude-skills --skill ecosystem -a claude-code. The skill will be placed in ~/.claude/skills/ecosystem/SKILL.md.

What does the ecosystem skill do?

It scans installed Claude Code skills for staleness, audits git repos in ~/ai_projects/ for instruction drift, and verifies services on a Mac Mini.

What paths does ecosystem expect by default?

It explicitly targets ~/.claude/skills/ for skills, ~/ai_projects/ for git repositories, and an SSH host named mac-mini.

Source text

SKILL.md · MIT
---
name: ecosystem
description: Audit the Claude Code ecosystem — skill health and staleness, project activity pulse, CLAUDE.md instruction drift, Mac Mini service status. Use this skill whenever the user asks about ecosystem health, stale skills, abandoned projects, system status, infrastructure check, "what's broken", "what's stale", "how's my setup", or any request to review the state of their Claude Code environment. Also triggers on "/ecosystem", "audit my ecosystem", "ecosystem health", "ecosystem audit".
---

# Ecosystem Audit

On-demand audit of the Claude Code ecosystem. Runs 4 checks, prints a full report with an attention summary, and appends a summary to today's daily note.

## Workflow

Run all 4 checks sequentially using the bash blocks below. Each block runs in a fresh shell, so after all checks complete, compose the daily note summary and attention rollup from the terminal output you've collected.

**CRITICAL**: Run `date +"%Y%m%d"` before writing to the daily note.

### Step 1: Skill Health

Scan `~/.claude/skills/` for skill freshness and broken symlinks.

```bash
NOW=$(date +%s)
D30=$((NOW - 30*86400))
D90=$((NOW - 90*86400))
ACTIVE="" BROKEN=""
ACTIVE_N=0 RECENT_N=0 STALE_N=0 BROKEN_N=0 TOTAL=0

for entry in ~/.claude/skills/*/; do
  [ -d "$entry" ] || continue
  name=$(basename "$entry")
  TOTAL=$((TOTAL + 1))

  if [ -L "${entry%/}" ] && [ ! -e "${entry%/}" ]; then
    target=$(readlink "${entry%/}")
    BROKEN="${BROKEN}  - ${name} -> ${target}\n"
    BROKEN_N=$((BROKEN_N + 1))
    continue
  fi

  real_path="$entry"
  if [ -L "${entry%/}" ]; then
    real_path="$(readlink "${entry%/}")/"
  fi

  newest=$(find "$real_path" -type f -exec stat -f %m {} + 2>/dev/null | sort -rn | head -1)
  [ -z "$newest" ] && newest=0

  if [ "$newest" -ge "$D30" ]; then
    ACTIVE="${ACTIVE}, ${name}"
    ACTIVE_N=$((ACTIVE_N + 1))
  elif [ "$newest" -ge "$D90" ]; then
    RECENT_N=$((RECENT_N + 1))
  else
    STALE_N=$((STALE_N + 1))
  fi
done

echo "### Skills (${TOTAL} total)"
echo "- Active (30d): ${ACTIVE_N}${ACTIVE:+ — ${ACTIVE:2}}"
echo "- Recent (30-90d): ${RECENT_N}"
echo "- Stale (>90d): ${STALE_N}"
echo "- Broken symlinks: ${BROKEN_N}"
[ -n "$BROKEN" ] && printf "$BROKEN"
```

### Step 2: Project Pulse

Scan `~/ai_projects/` for git repo activity and CLAUDE.md presence. Cap dormant and abandoned lists at 10 names to keep output readable.

```bash
NOW=$(date +%s)
D30=$((NOW - 30*86400))
D180=$((NOW - 180*86400))
ACTIVE="" DORMANT="" ABANDONED="" DORMANT_N=0 ABANDONED_N=0 NO_CLAUDE=0
ACTIVE_N=0 GIT_TOTAL=0 DIR_TOTAL=0 NODATA=0

for dir in ~/ai_projects/*/; do
  [ -d "$dir" ] || continue
  DIR_TOTAL=$((DIR_TOTAL + 1))
  [ -d "${dir}.git" ] || continue
  GIT_TOTAL=$((GIT_TOTAL + 1))

  last_commit=$(git -C "$dir" log -1 --format=%ct 2>/dev/null)
  if [ -z "$last_commit" ]; then
    NODATA=$((NODATA + 1))
    continue
  fi

  name=$(basename "$dir")

  if [ "$last_commit" -ge "$D30" ]; then
    ACTIVE="${ACTIVE}, ${name}"
    ACTIVE_N=$((ACTIVE_N + 1))
  elif [ "$last_commit" -ge "$D180" ]; then
    DORMANT_N=$((DORMANT_N + 1))
    [ "$DORMANT_N" -le 10 ] && DORMANT="${DORMANT}, ${name}"
  else
    ABANDONED_N=$((ABANDONED_N + 1))
    [ "$ABANDONED_N" -le 10 ] && ABANDONED="${ABANDONED}, ${name}"
  fi

  [ ! -f "${dir}CLAUDE.md" ] && NO_CLAUDE=$((NO_CLAUDE + 1))
done

DORMANT_SUFFIX=""
[ "$DORMANT_N" -gt 10 ] && DORMANT_SUFFIX=" and $((DORMANT_N - 10)) more"
ABANDONED_SUFFIX=""
[ "$ABANDONED_N" -gt 10 ] && ABANDONED_SUFFIX=" and $((ABANDONED_N - 10)) more"

echo ""
echo "### Projects (${DIR_TOTAL} dirs, ${GIT_TOTAL} git repos)"
echo "- Active (30d): ${ACTIVE_N}${ACTIVE:+ — ${ACTIVE:2}}"
echo "- Dormant (30-180d): ${DORMANT_N}${DORMANT:+ — ${DORMANT:2}${DORMANT_SUFFIX}}"
echo "- Abandoned (>6mo): ${ABANDONED_N}${ABANDONED:+ — ${ABANDONED:2}${ABANDONED_SUFFIX}}"
[ "$NODATA" -gt 0 ] && echo "- No data (empty/corrupt): ${NODATA}"
echo "- Missing CLAUDE.md: ${NO_CLAUDE}"
```

### Step 3: CLAUDE.md Drift

Check if CLAUDE.md files are stale relative to their project's latest commit. Uses git commit dates (not filesystem mtime) for accuracy.

```bash
NOW=$(date +%s)
STALE_LIST="" STALE_N=0
UNTRACKED_LIST="" UNTRACKED_N=0

while IFS= read -r claude_file; do
  project_dir=$(dirname "$claude_file")
  while [ ! -d "${project_dir}/.git" ] && [ "$project_dir" != "$HOME/ai_projects" ]; do
    project_dir=$(dirname "$project_dir")
  done
  [ -d "${project_dir}/.git" ] || continue

  claude_commit=$(git -C "$project_dir" log -1 --format=%ct -- "$claude_file" 2>/dev/null)
  project_commit=$(git -C "$project_dir" log -1 --format=%ct 2>/dev/null)
  [ -z "$project_commit" ] && continue

  if [ -z "$claude_commit" ]; then
    rel_path="${claude_file#$HOME/}"
    UNTRACKED_LIST="${UNTRACKED_LIST}  - ~/${rel_path}\n"
    UNTRACKED_N=$((UNTRACKED_N + 1))
    continue
  fi

  diff=$((project_commit - claude_commit))
  if [ "$diff" -gt $((90*86400)) ]; then
    rel_path="${claude_file#$HOME/}"
    claude_date=$(date -r "$claude_commit" +%Y-%m-%d)
    project_date=$(date -r "$project_commit" +%Y-%m-%d)
    STALE_LIST="${STALE_LIST}  - ~/${rel_path} (last: ${claude_date}, project: ${project_date})\n"
    STALE_N=$((STALE_N + 1))
  fi
done < <(find ~/ai_projects -maxdepth 2 -name CLAUDE.md 2>/dev/null)

echo ""
echo "### CLAUDE.md Drift"
echo "- Stale instructions (>90d behind project): ${STALE_N}"
[ -n "$STALE_LIST" ] && printf "$STALE_LIST"
echo "- Untracked: ${UNTRACKED_N}"
[ -n "$UNTRACKED_LIST" ] && printf "$UNTRACKED_LIST"
```

### Step 4: Mac Mini Health

SSH to agents-mac-mini and check services. Uses `-T` to avoid pseudo-tty and `ECOSYS:` prefix to filter iTerm escape codes from the output.

Custom services on the Mac Mini use various prefixes (`com.server.*`, `com.telegram-agent.*`, `com.photopulse.*`, `com.health.*`, `com.temporal.*`, `ai.hermes.*`, etc.), so count all non-Apple LaunchAgents.

```bash
echo ""
echo "### Mac Mini"

MINI_OUTPUT=$(ssh -T -o ConnectTimeout=5 -o BatchMode=yes mac-mini 'export TERM=dumb; thumb=$(curl -s --max-time 3 -o /dev/null -w "%{http_code}" http://localhost:8080/ 2>/dev/null); viz=$(curl -s --max-time 3 -o /dev/null -w "%{http_code}" http://localhost:8081/ 2>/dev/null); agents=$(launchctl list 2>/dev/null | tail -n +2 | awk "{print \$3}" | grep -cv "^com\.apple" || echo 0); hdb_size=$(stat -f %z ~/ai_projects/health-import/health.db 2>/dev/null || echo 0); echo "ECOSYS:${thumb}|${viz}|${agents}|${hdb_size}"' 2>/dev/null)

DATA=$(echo "$MINI_OUTPUT" | grep "^ECOSYS:" | sed 's/^ECOSYS://')

if [ -z "$DATA" ]; then
  echo "- Status: UNREACHABLE (SSH failed)"
else
  IFS='|' read -r THUMB VIZ AGENTS HDB_SIZE <<< "$DATA"
  [ "$THUMB" = "200" ] && echo "- Thumb server (:8080): UP" || echo "- Thumb server (:8080): DOWN (${THUMB})"
  [ "$VIZ" = "200" ] && echo "- Viz server (:8081): UP" || echo "- Viz server (:8081): DOWN (${VIZ})"
  echo "- Custom LaunchAgents: ${AGENTS}"
  if [ "$HDB_SIZE" -gt 0 ] 2>/dev/null; then
    HDB_GB=$(echo "scale=1; ${HDB_SIZE}/1073741824" | bc)
    echo "- Health DB: ${HDB_GB} GB"
  else
    echo "- Health DB: NOT FOUND"
  fi
fi
```

### Step 5: Attention Summary + Daily Note

After printing the full terminal report, add an attention summary highlighting anything that needs action:

```
### Attention
- [list any: broken symlinks, services DOWN, stale CLAUDE.md, Mac Mini unreachable]
- If nothing needs attention: "All clear."
```

Then compose a summary and write it to today's daily note.

Get today's date first:

```bash
TODAY=$(date +"%Y%m%d")
```

Build the summary block from the results you collected in Steps 1-4 (re-read the terminal output above — variables don't carry across bash blocks). Format:

```markdown
## Ecosystem

Skills: N active / N recent / N stale / N broken
Projects: N active / N dormant / N abandoned
CLAUDE.md drift: N stale, N untracked
Mac Mini: [status summary]

_Ecosystem audit · YYYY-MM-DD_
```

Write to `~/Brains/brain/Daily/${TODAY}.md`:
- If `## Ecosystem` section already exists, replace everything from `## Ecosystem` to the next `##` heading (or `- - -` separator)
- If it doesn't exist, insert above the first `- - -` separator
- If any red flags (broken symlinks > 0, services down, stale CLAUDE.md > 3), prepend `> [!warning] Ecosystem issues detected`

© glebis · MIT · glebis/claude-skills

Similar skills

All skills for Agent setup: memory, context, orchestration →

From the same set51

The whole set →
sorted Automates getSorted.de in a live browser to create freelancer invoices, track expenses, and file German tax reports. CX Codex +1 151installs rag-eval rag-eval evaluates RAG pipelines against a gold set using cost-capped sweep experiments. CX Codex +1 150installs wow-digest wow-digest collects newsletter emails and Telegram posts, selects surprising reads, and adds them to a daily note. Useful for a tailored morning reading list. CX Codex +1 145installs tg-responder tg-responder is a Claude Code skill for reviewing and approving Telegram reply drafts and tracking unanswered messages to manage your inbox and keep up with important conversations. CX Codex +1 145installs retrospective retrospective analyzes session logs, extracts corrections and workflow failures, then writes approved lessons directly into memory files, skills, or CLAUDE.md. CX Codex +1 135installs present present builds interactive HTML slide decks with synced ElevenLabs voiceovers, dual reading modes, and optional GPT Image 2 illustrations. CX Codex +1 117installs site-diagnosis site-diagnosis runs a pre-consultation questionnaire for AI-built websites and turns the answers into a consultation brief. Use it before a site review when the problems are still unclear. CX Codex +1 116installs learning-vault learning-vault builds a structured Markdown study vault for certifications or research topics. Generates atomic concepts, review tasks, and learning maps tailored to your confidence levels. CX Codex +1 108installs

Project sponsors

Companies that keep this analytics open