Menu
Load full context for modifying Claude Code configuration, settings, hooks, MCP servers, skills, or the wrapper script on this machine
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Create a new Jira ticket with user-approved summary and description
Assess the current state of a Jira ticket and produce a summary for development work
Convert an existing instruction file (scenario, agent, command) into a proper Claude Code skill
Claim a Jira ticket and set up tracking (administrative only - NO technical work)
Audit project-local Claude Code permissions and migrate appropriate ones to user scope
Expert code review of the most recent commit
Baseado na classificação ocupacional SOC
| name | x-claude-setup |
| description | Load full context for modifying Claude Code configuration, settings, hooks, MCP servers, skills, or the wrapper script on this machine |
| model | sonnet |
| disable-model-invocation | true |
| allowed-tools | Bash(*), Read, Write, Edit, Glob, Grep, Agent(claude-code-guide) |
Load comprehensive context for working on the Claude Code installation and configuration on this machine.
Input: $ARGUMENTS (optional: specific area to focus on, e.g. "hooks", "mcp", "permissions", "npm")
claude-code-guide AgentThis is not optional. Before you read any files, before you make any changes, before you do anything else: invoke the claude-code-guide agent. Every single time this skill is loaded, no exceptions. The user has repeatedly found that skipping this step leads to incorrect assumptions about Claude Code's configuration schemas, field names, valid values, and nesting levels. You will get things wrong without it.
Include this context in your prompt to the agent:
The user runs Claude Code on Linux Mint inside GNU screen + Kitty terminal. They have a Perl wrapper script, custom hooks, MCP servers, and a symlinked config structure. Please answer their question with full knowledge of Claude Code's capabilities.
Pass the user's specific question or task to the agent so it can provide targeted guidance.
Use this reference to understand the local installation before making changes.
~/.local/bin/claude (symlink to ~/.local/share/claude/versions/<version>)~/common/bin/claude (Perl) — the user invokes this, not the binary directly
~/common/conf/ai/devtools and ~/go/bin to $PATHscreen -Q info to propagate correct terminal dimensions ($COLUMNS/$LINES)$TERM=screen-256color and $CLAUDE_CODE_DISABLE_TERMINAL_TITLE=1CLAUDE.local.md for work repos using $CE_REMOTE_USERNAMEmsgwait 0/msgwait 3These files under ~/.claude/ are symlinks into ~/common/conf/ai/claude/ (version-controlled):
| Symlink | Target |
|---|---|
CLAUDE.md | conf/ai/claude/CLAUDE.md |
settings.json | conf/ai/claude/settings.json |
skills/ | conf/ai/claude/skills/ |
commands/ | conf/ai/claude/commands/ |
agents/ | conf/ai/claude/agents/ |
statusline-command.sh | conf/ai/claude/statusline-command.sh |
Not symlinked (local-only): .credentials.json, history.jsonl, sessions/, projects/, cache/, plugins/, plans/, tasks/
When editing config, edit the real file in ~/common/conf/ai/claude/ — the symlink ensures ~/.claude/ sees it. Changes are committable via git in ~/common.
Claude Code runs inside GNU screen sessions. This causes known issues:
screen -Q info fixaidoc/projects/home-network/summary:claude-code-terminal-corruption.mdsettings.json → hooks)| Event | Script | Purpose |
|---|---|---|
Notification | claude-code-set-notification | Writes JSON marker to /tmp/ when Claude needs attention |
UserPromptSubmit | claude-code-clear-notification | Clears marker when user provides input |
SessionEnd | claude-code-clear-notification | Clears marker on exit |
PostToolUse | claude-code-clear-notification | Clears marker after tool execution |
Not yet configured but available: PreToolUse hook with approve-safe-bash — a Perl script that auto-approves compound Bash commands (pipes, &&, ||) when all segments are read-only. See conf/ai/devtools/approve-safe-bash and aidoc/summary:approve-safe-bash-hook.md.
settings.json → mcpServers)| Server | Type | Notes |
|---|---|---|
google-sheets | Local node | ~/proj/claude-code-gsheets-mcp-server/google_sheets_mcp_server.js |
playwright | npx | @playwright/mcp@latest — runs via npx from cache |
atlassian-remote | SSE | https://mcp.atlassian.com/v1/sse |
~/.npmrc has ignore-scripts=true globally (defense against supply-chain attacks like the axios RAT incident)npm install --ignore-scripts=false <package>playwright MCP server runs via npx from a cached copy; reinstalling it requires the override flagsettings.json → permissions)The allow list is extensive — read-only commands, git read operations, skill scripts, MCP tools, and safe CLI tools are pre-approved. The deny list blocks destructive curl operations and dangerous find flags.
Key patterns:
/home/buddy/ and /home/bburden/ paths (home vs. work machine)git -C operations are allowed in settings but discouraged in CLAUDE.md when targeting cwdadditionalDirectories grants access to ~/common/aidoc, ~/common, and ~/docs/ai/screenshotsCustom bash script (statusline-command.sh) that parses the transcript to show:
~/common/conf/ai/devtools/)| Script | Purpose |
|---|---|
approve-safe-bash | PreToolUse hook: auto-approves read-only compound commands |
check-dir-age | Checks age of a directory (for staleness detection) |
claude-code-set-notification | Notification hook: writes attention-needed marker |
claude-code-clear-notification | Clears notification marker |
repo-last-pulled | Reports when a repo was last pulled |
With the above context loaded, help the user with their specific request. When modifying:
~/common/conf/ai/claude/settings.json (the real file)~/common/bin/claude~/common/conf/ai/claude/skills/~/common/conf/ai/devtools/~/common/conf/ai/claude/statusline-command.shOffer to commit changes when done.