cos-review
Skill19 starsChanged 8 months agoMITView on GitHub
Chief of Staff Review — diagnose, build, and grow your personal AI operating system. Three modes: setup (first run), review (bi-weekly check-in), status (quick health check).
- Installs packages
In this repository5 other files it gives its agents.
- atomizeSkill
- daily-logSkill
- linkedin-research-postSkill
- ai-personal-os-onboardingSkill
- slide-inspectorSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
79 rules. About 0 are common, 41 are distinctive, and 38 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈41
Distinctive · estimate
This author's own choices. Worth a look.
38
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 21
Check what the user typed:
- Distinctive: line 30
Run this the first time, or when rebuilding the system from scratch.
- Very specific: line 36
Important: Use Read, Glob, and Grep tools (cross-platform) instead of shell commands like find or grep.
- Very specific: line 39
Read ~/.claude/CLAUDE.md — exists?
- Very specific: line 41
Check ~/.claude/SOUL.md first (the standard global location)
- Very specific: line 42
If not there, Glob /SOUL.md in home (max depth 5, exclude node_modules/.git/Library)
- Very specific: line 43
If found elsewhere but NOT in ~/.claude/: report "SOUL.md exists at [path] but is not globally accessible.
- Very specific: line 44
Check if ~/.claude/CLAUDE.md references SOUL.md — if not, flag: "SOUL.md won't be auto-loaded unless CLAUDE.md references it"
- Very specific: line 45
Search for user-profile.md — same 4-step approach as SOUL.md
- Very specific: line 47
Check memory — use Glob ~/.claude/projects/*/memory/*.md and count entries
- Distinctive: line 60
If installed, check for both local models (e.g., qwen3.5) and cloud models (e.g., kimi-k2.5:cloud, qwen3.5:cloud)
- Very specific: line 62
If not installed and user has privacy or cost concerns, suggest: brew install ollama then ollama launch claude --model kimi-k2.5:cloud for …
- Very specific: line 64
If ActivityWatch IS running, check for the analysis skill: Glob ~/.claude/skills/activitywatch*/SKILL.md
- Distinctive: line 70
Check for key skills: morning, daily-log, atomize, council, cos-review
- Distinctive: line 71
If /morning skill is missing, check for EQUIVALENT morning systems:
- Very specific: line 72
Look for LaunchAgents/schtasks/cron with "morning" or "autopilot" (from L4 check)
- Very specific: line 73
Search common directories for morning scripts: ~/GH/*/morning*, ~/scripts/morning*
- Very specific: line 74
If found: "Your morning system exists at [path] as a [type].
- Very specific: line 79
Mac: check ~/Library/LaunchAgents/ for plist files matching morning/autopilot/claude/daily/briefing
- Very specific: line 80
Windows: run schtasks /query /fo LIST 2>NUL and search for morning/claude/daily/briefing entries
- Very specific: line 81
Linux: run crontab -l 2>/dev/null and search for morning/claude/daily entries
- Distinctive: line 84
If hooks exist, enumerate which event types are configured (e.g., SessionStart, PreCompact, PostToolUse, Stop)
- Very specific: line 85
Check that hook commands reference scripts that actually exist (if command points to a .sh file, verify it's there)
- Distinctive: line 86
If empty, proactively suggest two hooks:
- Distinctive: line 95
If only SessionStart exists, suggest PreCompact as the next hook
- Very specific: line 98
Check for a /heartbeat skill: Glob ~/.claude/skills/heartbeat/SKILL.md
- Distinctive: line 99
Check for a periodic LaunchAgent/cron matching "heartbeat" in addition to "morning"
- Very specific: line 100
Check ~/heartbeat.log for recent entries (if file exists, when was it last updated?)
- Distinctive: line 101
If no heartbeat exists, explain the concept and offer to build it: "Your CoS only wakes when you start a session or when /morning runs at 7 …
- Distinctive: line 103
If they want it, use the Guided Heartbeat Builder (see below)
- Very specific: line 105
Check for Telegram channel plugin (telegram@claude-plugins-official in enabledPlugins)
- Distinctive: line 106
Check for Claude Desktop Dispatch (ask the user)
- Very specific: line 107
If neither: suggest Telegram channels (/plugin install telegram@claude-plugins-official, then claude --channels) or Claude Desktop Dispatch …
- Distinctive: line 110
Check MULTIPLE paths for morning briefings (users store them differently):
- Very specific: line 113
Search for recent .md files with "morning" or "briefing" in the name
- Distinctive: line 114
If found: count files, find most recent, check if within last 3 days
- Distinctive: line 115
If not found: note it — this is the most common L5 gap
- Distinctive: line 119
Read the user's past Claude Code context to understand ACTUAL usage patterns.
- Very specific: line 121
Read ~/.claude/projects/ — list subdirectories (these are project contexts)
- Very specific: line 123
Look at ~/.claude/settings.json for recently used configuration
- Distinctive: line 124
Search recent memory entries for patterns:
- Very specific: line 159
Read references/diagnostic-questions.md for the full 25-question scoring criteria.
- Distinctive: line 167
Check understanding: "Do you know what [concept] is, or want me to explain first?" If the user is unsure, give a brief plain-language …
- Distinctive: line 172
Ask: "Want me to set this up now?
- Very specific: line 173
If yes, use the relevant guide from references/setup-guides/
- Very specific: line 177
Read references/layer-actions.md for specific actions per layer gap.
- Distinctive: line 181
If /morning is missing (one of the top priorities — it flips 3-5 questions across L3, L4, and L5), offer to build it through an interview:
- Distinctive: line 185
If yes, ask these questions ONE AT A TIME, waiting for each answer:
- Distinctive: line 188
Do you handle sensitive data that should stay local via Ollama?
- Distinctive: line 191
When this breaks at 7 AM, what do you check first?
- Very specific: line 193
After all answers, build the SKILL.md at ~/.claude/skills/morning/SKILL.md.
- Very specific: line 194
Use SOUL.md for voice if it exists.
- Distinctive: line 201
If the user has /morning working but no periodic awareness between sessions, offer to build a heartbeat:
- Distinctive: line 205
If yes, ask these questions ONE AT A TIME:
- Distinctive: line 208
When should it NOTIFY you vs stay silent?
- Distinctive: line 210
Should it auto-resolve safe items?
- Distinctive: line 212
After answers, build two things:
- Very specific: line 230
Test: run /heartbeat manually first.
- Distinctive: line 240
Only do 2-3 priorities per session.
- Distinctive: line 244
After setup is complete:
- Distinctive: line 248
If they have Telegram connected, offer to set a reminder: "Want me to send you a Telegram reminder in 2 weeks?"
- Distinctive: line 249
If yes, send a Telegram message with a reminder, or suggest they set a phone reminder/calendar event
- Very specific: line 259
Read the most recent file from ~/cos-reviews/.
- Very specific: line 263
Read references/diagnostic-questions.md for the 25 questions.
- Very specific: line 276
Check if /morning skill exists (~/.claude/skills/morning/SKILL.md) and when it last produced output (check ~/morning-briefings/ and …
- Very specific: line 277
Check memory entry count (Glob ~/.claude/projects/*/memory/*.md) — growing or stagnant?
- Very specific: line 278
Check Ollama: run ollama list
- Distinctive: line 279
Check scheduled tasks (platform-aware: LaunchAgents on Mac, schtasks on Windows, crontab on Linux)
- Very specific: line 280
Check hooks in ~/.claude/settings.json: which event types are configured?
- Very specific: line 281
Check heartbeat: does /heartbeat skill exist?
- Distinctive: line 299
Look for patterns in memory entries:
- Distinctive: line 312
If they name one → "That's your next skill.
- Distinctive: line 313
If blank → "Watch for repetitions this week.
- Distinctive: line 319
Not a list. ONE thing. The highest-leverage investment.
- Distinctive: line 375
No questions. No diagnostic.
- Very specific: line 377
Run all health checks from Step 3 of Review mode.
- Distinctive: line 407
Every complex system that works evolved from a simple system that worked.
- Distinctive: line 408
Do it manually. Do it again.
- Distinctive: line 420
Never overwhelming (ONE action, not ten)
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
/cos-review — Chief of Staff Review
Diagnose, build, and grow your personal AI operating system.
Your AI is your Chief of Staff. This skill is the performance review.
Detect Mode
Check what the user typed:
/cos-review setup→ run Setup Mode/cos-review status→ run Status Mode/cos-review(no arguments) → run Review Mode
MODE 1: Setup (first run)
Run this the first time, or when rebuilding the system from scratch. It scans everything, analyzes usage, and guides the user through fixing their biggest gaps.
Step 1: System Scan
Scan automatically — don't ask, CHECK. Report what exists and what's missing.
Important: Use Read, Glob, and Grep tools (cross-platform) instead of shell commands like find or grep. Only fall back to Bash when no dedicated tool works. Detect the OS first (uname on Mac/Linux, or check for Windows paths) and adapt commands accordingly.
L1: Knows You
- Read
~/.claude/CLAUDE.md— exists? word count? - Search for
SOUL.md:- Check
~/.claude/SOUL.mdfirst (the standard global location) - If not there, Glob
**/SOUL.mdin home (max depth 5, exclude node_modules/.git/Library) - If found elsewhere but NOT in
~/.claude/: report "SOUL.md exists at [path] but is not globally accessible. Recommend: copy to~/.claude/SOUL.md" - Check if
~/.claude/CLAUDE.mdreferences SOUL.md — if not, flag: "SOUL.md won't be auto-loaded unless CLAUDE.md references it"
- Check
- Search for
user-profile.md— same 4-step approach as SOUL.md - Count project-level CLAUDE.md files — use Glob
**/CLAUDE.mdfrom home (max depth 4, exclude node_modules/.git) - Check memory — use Glob
~/.claude/projects/*/memory/*.mdand count entries
L2: Has Sources
- MCPs and plugins are configured in THREE locations — check ALL of them:
~/.claude/.mcp.json— read and parse themcpServerskey (standalone MCP servers)~/.claude/settings.json— read and parse theenabledPluginsobject (key:boolean dict, e.g.,"gmail@claude-plugins-official": true)- Project-level
.mcp.jsonfiles — Glob for them in the current project
- For each source found, try a lightweight test call (skip if tool not available):
- Gmail: try
gmail_get_profileorgmail_list_labels - Telegram: try
list_chats(limit 1) - AnySite: try
duckduckgo_searchwith "test" - Calendar: try listing today's events
- Gmail: try
- Ollama: run
ollama list(works on Mac, Windows, Linux). If command not found, mark as not installed.- If installed, check for both local models (e.g., qwen3.5) and cloud models (e.g., kimi-k2.5:cloud, qwen3.5:cloud)
- Note: cloud models via Ollama are fast and cheap but data goes to a third-party provider. Local models are fully private but require a strong GPU — slow on laptops.
- If not installed and user has privacy or cost concerns, suggest:
brew install ollamathenollama launch claude --model kimi-k2.5:cloudfor cost savings, or local model for full privacy
- ActivityWatch (focus/productivity tracking): run
curl -s http://localhost:5600/api/0/info 2>/dev/nullvia Bash. If responds, it's running. If not, check ifaw-qtoraw-serverprocess exists. If installed but not running, note it. If not installed, suggest as optional: "ActivityWatch tracks your screen time and focus patterns — your /morning can use it to report deep work hours and app usage. Install from https://activitywatch.net"- If ActivityWatch IS running, check for the analysis skill: Glob
~/.claude/skills/activitywatch*/SKILL.md - If AW runs but no analysis skill: "You have ActivityWatch collecting data but no way to analyze it. Install the analysis skill from https://github.com/BayramAnnakov/activitywatch-analysis-skill — it calculates focus scores, detects death loops (repetitive app switching), and generates weekly insights. After installing, run it once with defaults then review uncategorized apps to calibrate for your workflow."
- If ActivityWatch IS running, check for the analysis skill: Glob
- Report which sources are installed, which respond, which have expired tokens
L3: Has Procedures
- List skills: Glob
~/.claude/skills/*/SKILL.md— count and list names - Check for key skills: morning, daily-log, atomize, council, cos-review
- If /morning skill is missing, check for EQUIVALENT morning systems:
- Look for LaunchAgents/schtasks/cron with "morning" or "autopilot" (from L4 check)
- Search common directories for morning scripts:
~/GH/*/morning*,~/scripts/morning* - If found: "Your morning system exists at [path] as a [type]. This works! Optionally create a /morning SKILL.md wrapper for consistency."
- Count total available slash commands (skills + built-in)
L4: Takes Initiative
- Detect platform, then check for scheduled tasks:
- Mac: check
~/Library/LaunchAgents/for plist files matching morning/autopilot/claude/daily/briefing - Windows: run
schtasks /query /fo LIST 2>NULand search for morning/claude/daily/briefing entries - Linux: run
crontab -l 2>/dev/nulland search for morning/claude/daily entries
- Mac: check
- Agent teams: read
~/.claude/settings.json, check ifenv.CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMSequals "1" (it's stored in theenvdict, NOT as a shell variable) - Hooks: read
~/.claude/settings.json, check ifhookskey has any entries (not just exists — check it's non-empty)- If hooks exist, enumerate which event types are configured (e.g., SessionStart, PreCompact, PostToolUse, Stop)
- Check that hook commands reference scripts that actually exist (if command points to a
.shfile, verify it's there) - If empty, proactively suggest two hooks:
- SessionStart — loads today's briefing into every session:
"SessionStart": [{"matcher": "", "hooks": [{"type": "command", "command": "cat ~/morning-briefings/$(date +%Y-%m-%d).md 2>/dev/null || echo 'No briefing today'", "timeout": 10}]}] - PreCompact — saves session transcripts before context compaction:
"PreCompact": [{"matcher": "", "hooks": [{"type": "command", "command": "your-transcript-backup-script.sh", "async": true, "timeout": 10}]}]
- SessionStart — loads today's briefing into every session:
- If only SessionStart exists, suggest PreCompact as the next hook
- Note: for complex hook logic, recommend external script files (e.g.,
~/.claude/hooks/session-start.sh) instead of inline commands
- Heartbeat (periodic sweep between sessions):
- Check for a
/heartbeatskill: Glob~/.claude/skills/heartbeat/SKILL.md - Check for a periodic LaunchAgent/cron matching "heartbeat" in addition to "morning"
- Check
~/heartbeat.logfor recent entries (if file exists, when was it last updated?) - If no heartbeat exists, explain the concept and offer to build it: "Your CoS only wakes when you start a session or when /morning runs at 7 AM. Between those moments, it's dormant — emails arrive, calendar changes, deadlines approach, and your CoS has no idea. A heartbeat is a silent periodic sweep (every 2-4 hours) that checks for changes and only pings you via Telegram if something needs attention. Think of it as the difference between checking your phone once a day vs having a smart notification system."
- If they want it, use the Guided Heartbeat Builder (see below)
- Check for a
- Two-way access: check if the user can reach their CoS from their phone
- Check for Telegram channel plugin (
telegram@claude-plugins-officialin enabledPlugins) - Check for Claude Desktop Dispatch (ask the user)
- If neither: suggest Telegram channels (
/plugin install telegram@claude-plugins-official, thenclaude --channels) or Claude Desktop Dispatch as options
- Check for Telegram channel plugin (
L5: Delivers Results
- Check MULTIPLE paths for morning briefings (users store them differently):
~/morning-briefings/— the recommended path from the course~/reports/daily/— common alternative- Search for recent
.mdfiles with "morning" or "briefing" in the name
- If found: count files, find most recent, check if within last 3 days
- If not found: note it — this is the most common L5 gap
Step 2: Conversation History Analysis
Read the user's past Claude Code context to understand ACTUAL usage patterns.
- Read
~/.claude/projects/— list subdirectories (these are project contexts) - For the 3-5 most recently modified projects, read their MEMORY.md
- Look at
~/.claude/settings.jsonfor recently used configuration - Search recent memory entries for patterns:
- What skills/commands are mentioned most?
- What MCPs appear in memory entries?
- What tasks or workflows are described?
- What frustrations or errors are noted?
Present findings:
USAGE PATTERNS (from your recent sessions):
- You use [skill X] frequently — strongest habit
- You use [MCP Y] in [N] different projects — active source
- You manually [task Z] multiple times — SKILL CANDIDATE
- [MCP A] is installed but never appears in sessions — installed ≠ used
- You haven't used [skill B] recently — dormantStep 3: System Map
Combine scan + conversation analysis into a visual report:
YOUR CHIEF OF STAFF — CURRENT STATE
Score: X/25 — [RANK]
L1: KNOWS YOU ████░ 4/5
L2: HAS SOURCES ███░░ 3/5
L3: HAS PROCEDURES ██░░░ 2/5
L4: TAKES INITIATIVE ░░░░░ 0/5
L5: DELIVERS RESULTS ░░░░░ 0/5
INSTALLED BUT UNUSED: [from conversation analysis]
MANUAL REPETITIONS: [potential skill candidates]
STRONGEST HABIT: [most-used skill/pattern]Read references/diagnostic-questions.md for the full 25-question scoring criteria.
Step 4: Setup Priority
Based on weakest layer, suggest a prioritized list. Fix BOTTOM-UP (L1 before L4).
For each priority:
- Explain WHY it matters (one sentence)
- Check understanding: "Do you know what [concept] is, or want me to explain first?" If the user is unsure, give a brief plain-language explanation before suggesting the fix. Examples:
- Hooks → "Hooks are automatic actions that fire at specific moments — like your CoS reading today's briefing before every conversation, without you asking."
- Agent teams → "Agent teams are multiple AI agents that talk to EACH OTHER and coordinate, unlike sub-agents which just report back to you."
- SOUL.md → "SOUL.md defines your AI's voice — how it writes, what tone it uses, what words it avoids. Without it, every output sounds generic."
- Ollama → "Ollama lets you run AI models on your own machine. Your data never leaves your laptop — useful for sensitive work."
- Ask: "Want me to set this up now? (y/n)"
- If yes, use the relevant guide from
references/setup-guides/ - Test it works
- Move to next priority
Read references/layer-actions.md for specific actions per layer gap.
Guided /morning Builder
If /morning is missing (one of the top priorities — it flips 3-5 questions across L3, L4, and L5), offer to build it through an interview:
"Your biggest gap is /morning. Want me to help you build it? I'll interview you."
If yes, ask these questions ONE AT A TIME, waiting for each answer:
- What data sources do you have? (Gmail, Calendar, AnySite, Vault, Telegram — list what works)
- Single agent, sub-agents, or agent team with Researcher/Analyst/Learner roles?
- Do you handle sensitive data that should stay local via Ollama?
- How should the briefing be delivered? (Telegram, terminal, saved to file)
- How should your system LEARN from yesterday's briefing?
- When this breaks at 7 AM, what do you check first?
After all answers, build the SKILL.md at ~/.claude/skills/morning/SKILL.md.
Use SOUL.md for voice if it exists. Save output to ~/morning-briefings/YYYY-MM-DD.md.
Include a Learner role or step that reads yesterday's briefing and feeds patterns forward.
Then test: run /morning and verify it produces output. Debug if needed.
Guided Heartbeat Builder
If the user has /morning working but no periodic awareness between sessions, offer to build a heartbeat:
"Your /morning runs at 7 AM. But between then and your next session, your CoS is dormant. Want me to build a /heartbeat that checks for changes every few hours and only pings you when something needs attention?"
If yes, ask these questions ONE AT A TIME:
- What should it check? (unread emails, calendar next 2 hours, approaching deadlines, task file changes)
- How often? (every 2 hours, every 4 hours, 3x/day at 10am/1pm/5pm)
- When should it NOTIFY you vs stay silent? (only urgent items? any new email from key people? calendar conflicts?)
- How should it notify? (Telegram message, save to ~/heartbeat.log, both)
- Should it auto-resolve safe items? (e.g., acknowledge meeting invites, file routine emails) Or just report?
After answers, build two things:
1. The /heartbeat skill at ~/.claude/skills/heartbeat/SKILL.md:
Steps:
1. SCAN — Check each data source (email count + urgent senders, calendar next 2h, task deadlines)
2. TRIAGE — For each item: urgent (notify now), routine (auto-resolve if allowed), ignorable (skip)
3. ACT — Auto-resolve safe items if user opted in (acknowledge, file, update task list)
4. NOTIFY — If anything needs human attention → send Telegram message with ONE sentence per item
5. LOG — Append timestamp + summary to ~/heartbeat.log
6. SILENT — If nothing changed: log "HEARTBEAT_OK", send no notification2. The scheduled trigger:
- Mac: create a LaunchAgent plist at
~/Library/LaunchAgents/com.user.heartbeat.plistthat runsclaude --print "Run /heartbeat"at the chosen intervals - Windows:
schtasks /create /tn "Heartbeat" /tr "claude --print \"Run /heartbeat\"" /sc daily /st 10:00(repeat for each time slot) - Linux: add crontab entries for each time slot
Test: run /heartbeat manually first. Verify it stays silent when nothing is urgent. Then enable the schedule.
Growth path to explain to the user:
Level 1: SessionStart hook — CoS loads context when YOU start talking (you have this)
Level 2: /morning at 7 AM — CoS prepares your day once (you have this)
Level 3: /heartbeat every 2-4h — CoS checks for changes, pings you only when needed (building now)
Level 4: Always-on Telegram — CoS listens 24/7, you can ask anytime from your phoneOnly do 2-3 priorities per session. "Good enough for today. Do the rest this week."
Step 5: Schedule First Review
After setup is complete:
- Save baseline to
~/cos-reviews/[DATE]-baseline.mdusing the YAML format below - Tell the user: "Your first review is in 2 weeks. Run
/cos-reviewthen." - If they have Telegram connected, offer to set a reminder: "Want me to send you a Telegram reminder in 2 weeks?"
- If yes, send a Telegram message with a reminder, or suggest they set a phone reminder/calendar event
MODE 2: Review (default — bi-weekly)
The regular performance review. Run every 2 weeks.
Step 1: Load Previous Review
Read the most recent file from ~/cos-reviews/. If none exists, say: "No previous review found. Run /cos-review setup first, or I'll do a fresh assessment now."
Step 2: Smart Diagnostic
Read references/diagnostic-questions.md for the 25 questions.
Don't re-ask everything. Be smart:
- Questions that were YES last time: spot-check 2-3 to verify (things can break)
- Questions that were NO last time: ask if anything changed
- This should take 2 minutes, not 10
Calculate new score + rank. Compare to previous.
Step 3: System Health Check
Actually TEST the system — don't ask, VERIFY:
- Discover MCPs from all 3 sources (
.mcp.json,enabledPlugins, project.mcp.json) — try a lightweight test call for each - Check if /morning skill exists (
~/.claude/skills/morning/SKILL.md) and when it last produced output (check~/morning-briefings/and~/reports/daily/) - Check memory entry count (Glob
~/.claude/projects/*/memory/*.md) — growing or stagnant? - Check Ollama: run
ollama list - Check scheduled tasks (platform-aware: LaunchAgents on Mac, schtasks on Windows, crontab on Linux)
- Check hooks in
~/.claude/settings.json: which event types are configured? If hook commands reference script files, verify they exist. - Check heartbeat: does
/heartbeatskill exist? Is it scheduled? Check~/heartbeat.logfor recent entries.
Report with status icons:
SYSTEM HEALTH
✅ Gmail MCP responding
⚠️ Calendar MCP token expired
✅ /morning last run: today 07:00
✅ Memory 24 entries (↑3 since last review)
⚠️ Vault last /atomize: 8 days ago
✅ Hooks SessionStart + PreCompact configured
✅ Heartbeat last sweep: 2h ago, HEARTBEAT_OKStep 4: Conversation Pattern Check
Scan MEMORY.md entries created since last review date (Glob ~/.claude/projects/*/memory/*.md, check file modification dates). Do NOT try to read conversation transcripts (.jsonl) — they are not accessible.
Look for patterns in memory entries:
- New skills mentioned?
- New MCPs referenced?
- Repeated tasks or complaints?
"Since your last review, your memory mentions [task] multiple times. That's a skill candidate."
Step 5: Growth Question
Ask ONE question:
"What task did you do manually 3+ times in the past 2 weeks?"
- If they name one → "That's your next skill. Want me to help you write it now?"
- If blank → "Watch for repetitions this week. Law #3: manual, twice, playbook."
Step 6: One Action
Based on weakest layer, suggest ONE specific action. Read references/layer-actions.md.
Not a list. ONE thing. The highest-leverage investment.
"Your ONE investment: [specific action]. That alone flips [N] of your Layer [X] gaps."
Offer to BUILD it: "Want me to do this now? (y/n)"
Step 7: Log
Save review to ~/cos-reviews/YYYY-MM-DD.md:
---
date: YYYY-MM-DD
score: X
rank: [Intern|Junior Assistant|Trained CoS|Senior CoS|Pravaya Ruka]
previous_score: X
delta: +/-X
layers:
L1: X
L2: X
L3: X
L4: X
L5: X
weakest: LX
health:
gmail: ok|expired|not_configured
calendar: ok|expired|not_configured
telegram: ok|not_configured
anysite: ok|not_configured
ollama: ok|not_installed
morning_skill: exists|missing
morning_last_run: YYYY-MM-DD|never
memory_entries: X
vault_last_atomize: YYYY-MM-DD|never
growth_candidate: "description of repeated task"
action: "the ONE action suggested"
---
## Review Notes
[Summary of the review conversation]
## Progress
[Score trend if 3+ reviews exist]Show the trend:
Apr 5: 14/25 JUNIOR ████████░░░░░
Apr 19: 17/25 TRAINED ██████████░░░
May 3: 19/25 TRAINED ███████████░░MODE 3: Status (quick health check)
No questions. No diagnostic. Just test everything and report. Takes 30 seconds.
Run all health checks from Step 3 of Review mode. Discover MCPs/plugins from all 3 config sources (.mcp.json, enabledPlugins, project .mcp.json). Test each. Report status + issues + fix commands.
Detect platform for scheduled task checks (Mac: LaunchAgents, Windows: schtasks, Linux: crontab).
YOUR CoS HEALTH — [DATE]
━━━━━━━━━━━━━━━━━━━━━━━━
✅ Gmail plugin responding
⚠️ Calendar plugin token expired 2d ago
✅ AnySite MCP responding
✅ Telegram MCP responding
✅ Ollama installed (qwen3.5 local + kimi-k2.5:cloud)
✅ /morning last run: today 07:00
✅ Memory 24 entries
⚠️ Vault last /atomize: 8 days ago
✅ Hooks SessionStart configured
✅ Heartbeat last sweep: 2h ago (HEARTBEAT_OK)
✅ Scheduled task LaunchAgent active (Mac)
ISSUES: 1 critical, 1 warning
FIX: Refresh Calendar token — run: /plugin install google-calendar@claude-plugins-officialThe 7 Laws (Context)
These principles guide the review. Read references/7-laws.md for full context.
- You fall to the level of your systems.
- Every complex system that works evolved from a simple system that worked.
- Do it manually. Do it again. Third time — write a playbook.
- Information flows up AND down. Without the downward arrows, it's a dead pipeline.
- Alignment over volume.
- Not "I will." My SYSTEM will.
- Your CoS is always briefed, knows every playbook, dispatches specialists, and never forgets.
Tone
- Direct, not preachy
- Celebrate progress ("Score went from 14 to 17 — that's real growth")
- Honest about gaps ("Layer 4 is still at zero. That's where the leverage is.")
- Action-oriented ("Want me to build this now?")
- Never overwhelming (ONE action, not ten)
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

