joogy06/agent-foundry/claude-code-cli

Claude Code CLI

Task-indexed reference for Claude Code 2.1.170 (and the 2.1.x line in general). Forked from the superpowers working-with-claude-code skill at v0.3.1, refreshed against local ground truth, and extended with custom files that document this machine's forge/bob/alf/pa/wiki ecosystem.

This SKILL.md is the navigation index. Detailed documentation lives in references/. Read only what the current task needs.

When to use

• Building anything that calls claude from the shell or a script (headless -p mode)
• Authoring or debugging Claude Code agents, skills, plugins, hooks, MCP servers
• Editing ~/.claude/settings.json, ~/.claude/CLAUDE.md, ~/.claude/AGENTS.md
• Cross-referencing Claude Code features against Antigravity CLI (agy), Codex CLI, or GitHub Copilot CLI (see antigravity-cli, gh-copilot-cli, cross-tool-portability sub-skill)
• Operating inside the local custom ecosystem (forge, bob, alf, pa, wiki, agent-teams)

Versions covered

When the local claude version drifts from this skill, run scripts/verify-claude-install.sh to flag stale sections.

Quick task index

Headless / scripted invocation

Interactive use

Agents and skills

Orchestration, teams, tasks, scheduling

Plugins and marketplaces

MCP servers

Hooks and settings

Memory, context, files

Authentication and providers

Cost, analytics, monitoring

Editor / IDE integrations

Compliance, security, networking

Custom local ecosystem (this machine only)

These are not documented in upstream superpowers — they are specific to this user's setup.

Permission modes (verified from claude --help)

Six modes, exactly:

--allow-dangerously-skip-permissions is a separate opt-in flag and should only be used in sandboxes with no internet access.

Top-level subcommands (verified)

agents       Manage background agents (Agent View dashboard; --json for scripting)
auth         Manage authentication (login/logout/status)
auto-mode    Inspect auto mode classifier configuration
doctor       Check Claude Code auto-updater health
install      Install Claude Code native build
mcp          Configure and manage MCP servers
              add | add-json | get | list | remove | serve | reset-project-choices
plugin       Manage Claude Code plugins
              install | uninstall | enable | disable | list | update | validate
              marketplace { add | list | remove | update }
project      Manage Claude Code project state (purge: delete transcripts/tasks/file history/config)
setup-token  Set up a long-lived authentication token
ultrareview  Run a cloud-hosted multi-agent code review of the current branch / a PR (--json, --timeout)
update       Check for updates and install if available

Verified against claude --help on 2.1.170 (2026-06-10). agents is the background-agent Agent View (see references/native-teams-tasks.md), not a list of configured subagents. Use claude agents --help for the current sub-options on this machine.

Headless (-p) cheat sheet

# One-shot prompt, plain text
claude -p "summarize this file" < notes.md
 
# JSON output for scripts
claude -p --output-format json "list TODOs in this repo"
 
# Strict structured output
claude -p --output-format json \
  --json-schema '{"type":"object","properties":{"todos":{"type":"array","items":{"type":"string"}}},"required":["todos"]}' \
  "list TODOs"
 
# Stream-json for incremental processing
claude -p --output-format stream-json --include-partial-messages "long task here"
 
# CI-safe minimal mode
ANTHROPIC_API_KEY=$KEY claude --bare -p "lint this file" \
  --add-dir . --settings ./ci-settings.json
 
# Pin a pre-defined agent for the call
claude -p --agent reviewer "review the diff in this branch"
 
# Inline custom agent
claude -p --agents '{"redactor":{"description":"Redacts secrets","prompt":"You are a secret redactor."}}' \
  --agent redactor "redact secrets from this file"
 
# Cap spend
claude -p --max-budget-usd 0.50 "research X and report"

See references/headless.md for the full surface.

Anti-patterns