server-doctor
Skill4 starsChanged 4 months agoMITView on GitHub
Use when auditing or repairing Linux or macOS hosts that run OpenClaw, Telegram bots, or related automation, especially when the operator needs to map hosts, runtimes, and safe access paths before diagnosis.
- Reads credentials
1 other project uses a version of this.
Adopted by
Most recently taken up by
- fablier-dev/claude-ops-skillstheir own version, first seen Oct 11, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- fablier-dev/claude-ops-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository43 other files it gives its agents.
- agent-browserSkill
- agent-introspectionSkill
- agentos-contentSkill
- carousel-instagramSkill
- chat-archiveSkill
- codex-imageSkill
- content-engineSkill
- cross-reviewSkill
- datawrapperSkill
- dev-pipelineSkill
- excalidrawSkill
- fast-loop-codingSkill
- groq-voiceSkill
- higgsfield-generateSkill
- higgsfield-marketplace-cardsSkill
- higgsfield-product-photoshootSkill
- higgsfield-soul-idSkill
- instagram-superpowerSkill
- learningsSkill
- loop-codingSkill
- markdown-newSkill
- mcp-api-buildSkill
- mcp-builderSkill
- memory-auditSkill
- miro-boardSkill
- perplexity-researchSkill
- presentSkill
- reel-radarSkill
- reels-analytics-for-brokersSkill
- reelsSkill
- remote-agent-desktopSkill
- security-auditSkill
- senior-brainstormSkill
- seo-tuneSkill
- telegram-chipSkill
- threads-contentSkill
- topic-monitorSkill
- transcriptSkill
- twitterSkill
- workshopSkill
- youtube-producerSkill
- youtube-thumbnailSkill
- yt-researchSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
29 rules. About 0 are common, 18 are distinctive, and 11 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈18
Distinctive · estimate
This author's own choices. Worth a look.
11
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Use this skill to inspect and stabilize hosts that run:
- Distinctive: line 23
If the operator cannot show where the bots live or cannot provide a safe way to inspect them, the skill may still help reason about likely …
- Distinctive: line 41
Do not jump from partial visibility or target uncertainty straight to outage language.
- Distinctive: line 61
Before running normal diagnostics, ask the operator for the minimum map needed to be useful.
- Distinctive: line 80
Use a compact intake like this:
- Distinctive: line 92
Do not demand that secrets be pasted into public chat.
- Distinctive: line 96
If the operator has not provided enough information to reach the environment, explicitly warn that the skill will be limited until access …
- Very specific: line 104
When in partial map, keep going with whatever is reachable, but always list the missing facts blocking reliable repair.
- Distinctive: line 110
ask for the next highest-value access detail that would unlock diagnosis or repair
- Distinctive: line 114
Start with low-risk inspection:
- Distinctive: line 142
If the task is local rather than remote, adapt the same checks to the local shell.
- Distinctive: line 146
Check for service managers and containers.
- Distinctive: line 192
When a service interacts with Telegram, document:
- Distinctive: line 204
If a Telegram username is confirmed, record it only in the operator's private inventory unless the user explicitly wants a public-safe …
- Distinctive: line 235
Never publish passwords, tokens, chat IDs, session strings, private SSH config, or provider API keys.
- Distinctive: line 236
Never require the operator to paste secrets into the skill document itself.
- Very specific: line 237
Do not copy .env, openclaw.json, service unit files, or bot configs verbatim into public docs.
- Very specific: line 246
Prefer phrases like main OpenClaw gateway, content publishing bot, maintenance bot, and private Telegram channel.
- Very specific: line 254
read references/health-claims-and-evidence.md before making strong health or outage claims
- Very specific: line 255
read references/outage-classification.md when you need to label health state, degradation, recovery, or escalation level
- Distinctive: line 273
document impact, recovery, and remaining unknowns
- Very specific: line 283
Read references/openclaw-incident-response.md for concrete public-safe scenarios including:
- Very specific: line 304
Prefer a local Markdown mirror of https://docs.openclaw.ai before falling back to live browsing when the repo checkout is available.
- Very specific: line 321
if tools/chip-docs-local is missing, say that the local docs toolchain is unavailable instead of pretending the mirror exists
- Very specific: line 322
if dependencies are not installed in the submodule, run npm --prefix tools/chip-docs-local install before the sync command
- Distinctive: line 323
use the local mirror as the default documentation source during diagnosis when it is present and fresh enough for the task
- Very specific: line 324
check references/openclaw-docs/FILELIST.md first for page discovery
- Very specific: line 325
if the mirror is missing or stale and local shell access exists, refresh it with npm run sync-openclaw-docs
- Distinctive: line 326
if the refresh fails, say so plainly and fall back to live public docs
≈ “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
Server Doctor
Use this skill to inspect and stabilize hosts that run:
- OpenClaw gateways
- Telegram bots
- background automation
- Dockerized bot services
- user-space processes
systemdorlaunchdservices
This public version is intentionally generic. It does not assume any specific hostnames, users, IPs, bot usernames, chat IDs, directories, or credentials.
Non-negotiable warning
This skill is severely limited without access to the relevant bots, servers, containers, or local project directories.
If the operator cannot show where the bots live or cannot provide a safe way to inspect them, the skill may still help reason about likely causes, but it will be partially blind and often operationally weak. Say that plainly up front.
The skill should therefore start by building an environment map and an access map before attempting deeper diagnosis or repair.
Operational diagnosis must separate:
- spec correctness:
- correct host, runtime, owner, and service target
- correct source of truth
- current direct evidence
- wording that matches only what the evidence proves
- ops quality:
- healthy
- degraded
- partial failure
- unstable
- down
Do not jump from partial visibility or target uncertainty straight to outage language.
Primary goals
- Build a usable map of hosts, local machines, bots, runtimes, and safe access paths.
- Identify what is actually running on each reachable machine.
- Determine which Unix user owns each bot or service.
- Classify runtime style:
systemdlaunchd- Docker / Compose
- direct user-space process
- terminal multiplexer session
- Locate logs, configs, restart paths, and dependencies.
- Capture operational risks without leaking secrets.
Workflow
1. Access & Inventory Preflight
Before running normal diagnostics, ask the operator for the minimum map needed to be useful.
Collect whatever is already known:
- which servers, VPSes, Macs, PCs, NAS devices, or local laptops are in scope
- which bots or automations are believed to run on each machine
- which Unix user, container, service account, or local profile owns each runtime
- any known working directories, Compose projects, unit names, launch agents, cron jobs, or repo checkouts
- how the agent can safely reach each target:
sshtailscale- local shell
docker execkubectl execsystemctl --user- terminal multiplexer
- jump host / bastion
- operator-provided command output only
Use a compact intake like this:
Host or machine:
Role:
Known bots/services:
Runtime owner:
Known paths or service names:
Safe access method:
Missing information:Do not demand that secrets be pasted into public chat. The requirement is access, not credential leakage. Work with the operator to establish a safe connection path.
2. Declare map status
If the operator has not provided enough information to reach the environment, explicitly warn that the skill will be limited until access improves.
Track one of these states:
mapped- host, runtime, and access path are knownpartial map- some facts are known, but ownership, location, or access is missingunreachable- the target exists, but there is no current safe access path
When in partial map, keep going with whatever is reachable, but always list the missing facts blocking reliable repair.
In partial map mode:
- record confirmed facts separately from assumptions
- list unknowns explicitly
- ask for the next highest-value access detail that would unlock diagnosis or repair
3. Establish host context
Start with low-risk inspection:
hostname
whoami
uname -a
uptime
pwdOn Linux:
df -h
free -h
ps -eo user,pid,ppid,cmd --sort=user
ss -tulpnOn macOS:
sw_vers
df -h
ps aux
launchctl listIf the task is local rather than remote, adapt the same checks to the local shell.
4. Discover runtimes
Check for service managers and containers.
Linux:
systemctl --failed
systemctl list-units --type=service --all
docker ps -a
docker compose lsmacOS:
launchctl list | grep -Ei 'openclaw|bot|agent|telegram'
ls -la ~/Library/LaunchAgentsMultiplexers and direct processes:
tmux ls
screen -ls
ps aux | grep -Ei 'openclaw|telegram|bot|node|python' | grep -v grep5. Locate project state and bot directories
Typical areas to inspect:
~/.claude-lab(agent workspaces)~/workspace/opt/...- project directories containing
docker-compose.yml,compose.yaml,package.json,pyproject.toml,requirements.txt
Useful searches:
find ~ -maxdepth 3 -type d | grep -Ei 'openclaw|bot|telegram'
find /opt -maxdepth 4 -type f 2>/dev/null | grep -Ei 'openclaw|compose|docker|telegram'Also inspect any operator-provided checkout or local path directly if the bot is run from a workstation instead of a server.
6. Confirm Telegram-facing services
When a service interacts with Telegram, document:
- host or local machine
- owning Unix user
- runtime style
- process or service name
- working directory
- safe access path used to reach it
- log path
- restart command
- whether a Telegram bot username is confirmed
If a Telegram username is confirmed, record it only in the operator's private inventory unless the user explicitly wants a public-safe example.
7. Build the operational map
For each discovered service, capture:
- host role
- host or local machine name
- user
- service name
- bot name or function
- runtime
- access method
- startup mechanism
- function
- dependencies
- known failure modes
- safe operator commands for status/logs/restart
- unknowns still blocking deeper work
The map should make it obvious:
- where each bot lives
- how to connect to it safely
- what starts it
- where logs live
- how to restart it
- what is still unknown
Documentation rules
- Never publish passwords, tokens, chat IDs, session strings, private SSH config, or provider API keys.
- Never require the operator to paste secrets into the skill document itself.
- Do not copy
.env,openclaw.json, service unit files, or bot configs verbatim into public docs. - Redact:
- IP addresses
- domains
- email addresses
- Telegram usernames
- internal bot names
- hostnames
- user names if they identify private environments
- Prefer phrases like
main OpenClaw gateway,content publishing bot,maintenance bot, andprivate Telegram channel.
Incident handling
For OpenClaw or Telegram incidents:
- confirm which host or local machine owns the failing runtime
- confirm that a safe access path actually exists
- read
references/health-claims-and-evidence.mdbefore making strong health or outage claims - read
references/outage-classification.mdwhen you need to label health state, degradation, recovery, or escalation level - confirm the canonical runtime target before strong claims:
- host
- owner
- unit, launch agent, container, or process tree
- runtime directory or state directory
- live port, socket, or endpoint
- confirm whether the main process is running
- inspect recent logs
- classify the failure before changing anything major:
- process down
- Telegram transport broken
- auth or model fallback
- queue starvation or provider delay
- bootstrap bloat or startup tax
- dependency failure
- identify whether failure is transport, auth, upstream model, Telegram delivery, dependency, or operator-access related
- verify restart path
- document impact, recovery, and remaining unknowns
Claim discipline:
- incomplete visibility is not outage proof
- direct live probe beats stale legacy checks unless a stronger contradiction appears
- wrong runtime path, old port, duplicate runtime, or mirror copy must be ruled out before outage wording
- degraded, partial failure, unstable, and down are distinct labels
- restart alone is not recovery
Read references/openclaw-incident-response.md for concrete public-safe scenarios including:
- stale session model override
- per-agent auth-profile drift
- post-update Telegram transport regression
- bootstrap-bloat and startup-tax
- duplicate OpenClaw runtime on macOS
Public helper scripts in this repo:
scripts/normalize-openclaw-models.py- normalize model defaults and optional per-agent bindings
scripts/openclaw-auth-profile-sync.sh- inspect, sync, and validate per-agent auth-profile drift for a chosen profile id
scripts/openclaw-telegram-transport-hotfix.sh- re-apply a host-local Telegram transport compatibility patch after updates
scripts/macos-single-openclaw-runtime.sh- collapse a macOS host back to one canonical OpenClaw runtime
Public OpenClaw docs mirror
Prefer a local Markdown mirror of https://docs.openclaw.ai before falling back to live browsing when the repo checkout is available.
Tooling:
- submodule:
tools/chip-docs-local - source manifest:
sources/openclaw.yaml - refresh command from the repo root:
npm run sync-openclaw-docs - materialized snapshot root:
references/openclaw-docs/
Expected mirror artifacts:
references/openclaw-docs/current/references/openclaw-docs/FILELIST.mdreferences/openclaw-docs/state.json
Operating rules:
- if
tools/chip-docs-localis missing, say that the local docs toolchain is unavailable instead of pretending the mirror exists - if dependencies are not installed in the submodule, run
npm --prefix tools/chip-docs-local installbefore the sync command - use the local mirror as the default documentation source during diagnosis when it is present and fresh enough for the task
- check
references/openclaw-docs/FILELIST.mdfirst for page discovery - if the mirror is missing or stale and local shell access exists, refresh it with
npm run sync-openclaw-docs - if the refresh fails, say so plainly and fall back to live public docs
Load these references when needed:
references/openclaw-host-audit.mdreferences/health-claims-and-evidence.mdreferences/outage-classification.mdreferences/openclaw-incident-response.md
Output expectations
A good output from this skill should give the operator:
- a host-by-host or machine-by-machine map
- a bot-to-host map
- a user-by-user runtime inventory
- a runtime map showing how each bot is started
- an access map showing how each reachable target can be inspected safely
- the role of each bot
- restart/log commands
- current risks
- known unknowns that still need deeper access
- a clear statement when the work is being done in
partial mapmode
The result should be readable, operationally useful, and safe to share publicly.
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.

