cli-claude-code

Claude Code CLI Orchestrator - Cross-AI Task Delegation

CRITICAL — SELF-INVOCATION PROHIBITED

This skill dispatches to the Anthropic CLI binary (claude). If the agent currently reading this skill is itself running inside Claude Code (detection signals listed in §2), the skill MUST refuse to load and return the documented error message instead of generating any claude invocation.

A running CLI skill never dispatches itself. The cli-X skills are for cross-AI delegation only — never self-invocation.

Orchestrate Anthropic's Claude Code CLI from external AI assistants (OpenCode, Copilot, etc.) for tasks that benefit from deep extended thinking, surgical code editing, structured output with JSON schema validation, agent delegation, or persistent memory context.

Core Principle: The calling AI stays the conductor. Delegate to Claude Code for what it does best — deep reasoning, precise code editing, and structured analysis. Validate and integrate the output.

1. WHEN TO USE

Activation Triggers

• Deep Reasoning — extended-thinking architectural decisions, multi-dimensional trade-off analysis, step-by-step algorithm design, root-cause analysis of subtle bugs.
• Code Editing — surgical diff-based edits, pattern-preserving refactors, and multi-file coordinated changes.
• Structured Output — --json-schema-validated output, machine-readable analysis, guaranteed-structure data extraction, pipeline integration.
• Code Review — second-AI security audits, extended-thinking architecture review, cross-AI validation, pre-merge quality gates.
• Agent-Delegated Tasks — specialized .claude/agents/*.md matches, --permission-mode plan read-only exploration, @ai-council planning, session continuity (--continue, --resume).
• Background Processing — long-running offloaded analysis, parallel generation/docs, batch processing with --max-budget-usd cost control.

When NOT to Use

• You ARE Claude Code already. If your runtime is Claude Code (detection signal: $CLAUDECODE env var set, claude in process ancestry, or ~/.claude/state/<id>/lock present), this skill refuses to load. Self-invocation creates a circular dispatch loop and burns tokens for no value. The cli-X family is exclusively for cross-AI delegation.
• Simple, quick tasks where CLI overhead is not worth it.
• Tasks requiring interactive terminal UI (use claude directly instead).
• Context already loaded and understood by the calling AI.
• Tasks where Claude Code CLI is not installed.
• Real-time web search (Claude Code has no --search flag — use OpenCode).

2. SMART ROUTING

Prerequisite Detection

# Verify Claude Code CLI is available before routing
command -v claude || echo "Not installed. Run: npm install -g @anthropic-ai/claude-code"

# SELF-INVOCATION GUARD: If you ARE Claude Code, do not use this skill — use native capabilities
[ -n "$CLAUDECODE" ] && echo "ERROR: Already inside Claude Code session. Do not self-invoke."

Self-Invocation Guard

def detect_self_invocation():
    """Returns a non-None signal when the orchestrator is already running inside Claude Code."""
    # Layer 1: env var lookup — Claude Code sets CLAUDECODE on session start
    if os.environ.get('CLAUDECODE'):
        return ('env', 'CLAUDECODE')
    # Layer 2: process ancestry — claude in parent tree
    try:
        ancestry = subprocess.check_output(['ps', '-o', 'command=', '-p', str(os.getppid())]).decode()
        if '/claude' in ancestry or 'claude ' in ancestry:
            return ('ancestry', 'claude')
    except subprocess.SubprocessError:
        pass
    # Layer 3: state lock-file probe
    state_dir = os.path.expanduser('~/.claude/state')
    if os.path.isdir(state_dir):
        for entry in os.listdir(state_dir):
            if os.path.exists(os.path.join(state_dir, entry, 'lock')):
                return ('lockfile', entry)
    return None

if detect_self_invocation():
    refuse(
        "Self-invocation refused: this agent is already running inside Claude Code. "
        "Use a sibling cli-* skill or a fresh shell session in a different runtime to dispatch a different model."
    )

Resource Loading Levels

Smart Router

Provider-specific dictionaries (used by the shared helper functions in system-spec-kit/references/cli/shared-smart-router.md):

INTENT_SIGNALS = {
    "DEEP_REASONING":    {"weight": 4, "keywords": ["reason", "think", "analyze", "trade-off", "architecture", "extended thinking", "chain-of-thought", "root cause", "root-cause", "figure out why", "weigh the options", "pros and cons", "algorithm design", "architectural", "chain of thought", "extended-thinking", "step by step", "deep dive", "reasoning", "thinking through", "analysis", "trade-offs", "in-depth", "thorough", "carefully consider", "dig into", "get to the bottom of", "understand why", "evaluate the options", "unpack this", "difficult decision", "hard choice", "which direction", "long-term implications", "multiple factors", "nuanced", "compare and contrast", "make the call", "downstream effects", "well-thought-out"]},
    "CODE_EDITING":      {"weight": 4, "keywords": ["edit", "refactor", "modify", "fix", "change code", "surgical edit", "diff-based", "update the code", "rewrite", "restructure", "rename", "patch", "clean up", "rework", "multi-file edit", "diff based", "preserve existing patterns", "coordinated changes"]},
    "STRUCTURED_OUTPUT": {"weight": 4, "keywords": ["json", "schema", "structured", "extract", "parse", "validate output", "--json-schema", "machine-readable", "typed output", "well-formed", "extraction", "guaranteed structure", "pipeline integration", "return format", "field mapping"]},
    "REVIEW":            {"weight": 4, "keywords": ["review", "audit", "security", "quality", "second opinion", "cross-validate", "sanity check", "vulnerability", "pre-merge", "double-check", "look over", "flag concerns", "catch bugs", "spot issues"]},
    "AGENT_DELEGATION":  {"weight": 4, "keywords": ["delegate", "agent", "background", "parallel", "offload", "claude agent", "hand off", "farm out", "dispatch to", "read-only exploration", "plan mode", "spin off", "run independently"]},
    "TEMPLATES":         {"weight": 3, "keywords": ["template", "prompt", "how to ask", "claude prompt", "how do I phrase", "how should I word", "phrasing", "wording", "boilerplate"]},
    "PATTERNS":          {"weight": 3, "keywords": ["pattern", "workflow", "orchestrate", "session", "continue", "resume", "orchestration", "cross-ai", "conversation history", "pick up where we left off", "carry over context", "multi-step flow"]},
    # WHY: DESIGN is a deliberate cross-skill handoff, not an in-skill resource intent — when design
    # keywords fire, this skill's job is to route the work AWAY to sk-design-md-generator rather than
    # load local markdown. RESOURCE_MAP intentionally has no DESIGN entry and never will; the durable
    # sk-design-md-generator loading contract lives in the "Design Standards Loading" rule and the
    # DESIGN_DISPATCH_MANIFEST rule (Section 4 RULES), not in a same-skill reference file.
    "DESIGN":            {"weight": 4, "keywords": ["sk-design-md-generator", "extract design system", "generate design.md", "style reference", "design tokens", "css extraction", "tokens.json"]},
}

RESOURCE_MAP = {
    "DEEP_REASONING":    ["references/cli-reference.md", "references/claude-tools.md"],
    "CODE_EDITING":      ["references/cli-reference.md", "assets/prompt-templates.md"],
    "STRUCTURED_OUTPUT": ["references/cli-reference.md", "references/claude-tools.md"],
    "REVIEW":            ["references/integration-patterns.md", "references/agent-delegation.md"],
    "AGENT_DELEGATION":  ["references/agent-delegation.md", "references/integration-patterns.md"],
    "TEMPLATES":         ["assets/prompt-templates.md", "references/cli-reference.md"],
    "PATTERNS":          ["references/integration-patterns.md", "references/cli-reference.md"],
}

LOADING_LEVELS = {
    "ALWAYS": ["references/cli-reference.md", "assets/prompt-quality-card.md"],
    "ON_DEMAND_KEYWORDS": ["full reference", "all templates", "deep dive", "complete guide", "extended thinking", "json schema", "claude agent", "claude prompt", "diff-based edit"],
    "ON_DEMAND": ["references/claude-tools.md", "assets/prompt-templates.md"],
}

UNKNOWN_FALLBACK_CHECKLIST = [
    "Is the user asking about Claude Code CLI specifically?",
    "Does the task benefit from deep reasoning or extended thinking?",
    "Is structured JSON output needed (--json-schema)?",
    "Would surgical code editing or agent delegation help?",
]

Call sequence (using shared helpers from shared-smart-router.md):

1. discover_markdown_resources() — recursively enumerate current .md files under existing references/ and assets/ folders at routing time.
2. _guard_in_skill() + load_if_available() — sandbox paths to this skill, reject non-markdown loads, skip missing files, and suppress duplicates.
3. score_intents(task) and select_intents(scores, ambiguity_delta=1.0) — preserve provider-specific weighted intent scoring and top-2 ambiguity handling.
4. get_routing_key(task, intents) — derive the provider routing key from task/provider context, then fall back to claude_code.
5. ALWAYS-load LOADING_LEVELS["ALWAYS"], then return UNKNOWN_FALLBACK with UNKNOWN_FALLBACK_CHECKLIST when max score is 0.
6. CONDITIONAL-load RESOURCE_MAP[intent], ON_DEMAND-load keyword matches, and return a notice when no provider-specific knowledge base is available beyond always-load resources.

The route_claude_code_resources(task) function body lives in shared-smart-router.md — substitute <PROVIDER> = claude_code.

3. HOW IT WORKS

Prerequisites

# Verify installation
command -v claude || echo "Not installed. Run: npm install -g @anthropic-ai/claude-code"

# Self-invocation guard
[ -n "$CLAUDECODE" ] && echo "ERROR: Already inside a Claude Code session — do not self-invoke"

# Authentication — OAuth (Claude subscription), no API key
claude auth login        # interactive OAuth (browser flow)
# claude setup-token     # non-interactive OAuth token for CI/CD

Authentication: cli-claude-code authenticates through the Claude subscription OAuth only — claude auth login (interactive browser flow) or claude setup-token (a non-interactive OAuth token for CI/CD). It does not use an ANTHROPIC_API_KEY.

Provider Auth Pre-Flight (Claude Subscription OAuth)

MANDATORY before any first dispatch in a session. cli-claude-code authenticates through the Claude subscription OAuth only. If neither claude auth login nor a claude setup-token session is configured on this machine, a dispatch fails with 401 Unauthorized mid-round-trip. Run this check once per session, cache the result, and re-run it only if a dispatch fails with an auth error.

# One-shot pre-flight: capture OAuth status for routing
CLAUDE_AUTH=$(claude auth status 2>&1)
echo "$CLAUDE_AUTH" | grep -qi "authenticated\|logged in\|oauth\|setup-token" && CLAUDE_OAUTH_OK=1 || CLAUDE_OAUTH_OK=0

Decision tree (apply in order — first match wins):

User prompt template — not authenticated:

Claude Code is not authenticated on this machine. cli-claude-code uses Claude subscription OAuth only.
Run one, then confirm — the skill will retry the original dispatch:
  - `claude auth login`     (interactive browser OAuth flow)
  - `claude setup-token`    (non-interactive OAuth token for CI/CD)

Error-recovery contract. If a dispatch returns an auth error after pre-flight passed (OAuth expired or revoked), invalidate the cache, re-run claude auth login, and re-check before retrying. Never substitute a model the user didn't approve.

Default Invocation (Skill Default)

Default model + flags + agent: claude-sonnet-4-6 · --output-format text · no --agent (general-purpose). For deep-reasoning work, override with --model claude-opus-4-6 --effort high. The pinned shape:

claude -p "<prompt>" \
  --model claude-sonnet-4-6 \
  --output-format text \
  2>&1

User override (honor explicit user phrasing verbatim):

Model Selection

claude-sonnet-4-6 is the skill default. Reach for claude-opus-4-6 (deep reasoning / complex architecture — pair with --effort high) or claude-haiku-4-5-20251001 (fast, lightweight; only when explicitly requested); the current-generation claude-opus-4-8 / claude-sonnet-5 / claude-fable-5 IDs are selectable by name where the environment supports them. Full roster with tiers, cost, defaults, and the --effort mapping → references/providers-and-models.md.

Claude Code Agent Delegation

Route to a specialized .claude/agents/*.md agent with --agent <name> when the task matches a specialization. Full roster and invocation patterns: agent-delegation.md.

Dispatch-Critical Gotchas

The full flag glossary, unique capabilities (--json-schema, --max-budget-usd, extended thinking, session --continue/--resume), essential command examples, and troubleshooting table are in the ALWAYS-loaded cli-reference.md (§4–§13). Four gotchas that must be honored at routing time:

• Non-interactive requires -p (print) mode — claude -p "prompt" --output-format text 2>&1. --output-format defaults to text; use json (adds role/content/cost metadata) or stream-json only when a pipeline needs it. Capture stderr with 2>&1.
• --permission-mode plan is read-only — use it for review/analysis/exploration (no file writes). bypassPermissions auto-approves all writes and requires explicit user approval; the default mode already allows writes.
• No --search flag — Claude Code has no live web browsing. Route web research to cli-opencode.
• Check $CLAUDECODE before dispatch — a set value means the caller is already inside Claude Code; refuse (self-invocation), do not dispatch.

4. RULES

✅ ALWAYS

1. Verify Claude Code CLI is installed before first invocation (command -v claude); check $CLAUDECODE for nesting.

2. Use --permission-mode plan for review/analysis/exploration (no file writes); --output-format text unless JSON is specifically needed.

3. Validate output before applying — correctness, completeness, alignment, syntax checks if code generated.

4. Capture stderr (2>&1) to catch errors and warnings.

5. Specify --model explicitly: default claude-sonnet-4-6 unless task needs Opus (deep reasoning). Use Haiku only when explicitly requested or after adoption.

6. Route to the appropriate --agent <name> when the task matches a specialization (see Section 3 routing table).

7. Pass the spec folder to the delegated agent in the prompt: if the calling AI has an active Gate-3 spec folder, include Spec folder: <path> (pre-approved, skip Gate 3). If none, ASK the user before delegating — the delegated agent cannot answer Gate 3 interactively.

8. Prompt construction & model-craft (cli- family precedence).* Compose every dispatch prompt via the 3-tier rule canonical in ../../sk-prompt/sk-prompt-models/assets/cli-prompt-quality-card.md:

   1. Fast path (default). Build from the local assets/prompt-quality-card.md, which delegates the framework table + CLEAR check to the canonical card.
   2. Model override (mandatory for a profiled model). If the target model has a profile at ../../sk-prompt/sk-prompt-models/references/models/<id>.md, that profile OVERRIDES the cross-model default. The sk-prompt/sk-prompt-models hub owns per-model prompt-craft (framework + scaffold + gotchas, mirroring sk-prompt/sk-prompt-models/assets/model-profiles.json recommended_frameworks); consult it before composing for any small model.
   3. Deep path (escalation). Dispatch @prompt-improver via the Task tool (never load full sk-prompt inline) when any canonical Tier 3 trigger applies — the trigger list lives in ../../sk-prompt/sk-prompt-models/assets/cli-prompt-quality-card.md under "Tier 3 — Deep path"; do not re-enumerate it here.

   Tag the framework in the Bash invocation comment and use the returned ENHANCED_PROMPT. Apply the CLEAR 5-question check from the canonical card via the local delegating card.

9. Code Standards Loading (surface-aware contract) — When dispatching for code review or code generation, instruct the dispatched session to: (1) load sk-code; (2) let sk-code emit a surface tag matching the detected stack from markers and target files; (3) load the selected surface resources and run its verification commands; (4) load sk-code's code-review mode only for formal findings-first review output. Fallback: if the surface cannot be determined confidently, ask for the runtime surface and verification command set. NEVER hardcode obsolete sibling code skills in dispatch prompts.

10. Design Standards Loading (measured-reference contract) — When dispatching for design or UI work, instruct the dispatched session to: (1) load sk-design-md-generator; (2) extract a measured Style Reference DESIGN.md (named color tokens, type scale, components, Quick-Start CSS/Tailwind) from the live source before building UI; (3) build against those measured tokens and run the extraction's validate step to confirm hex/section fidelity. Fallback: if there is no live source to measure, ask for the reference URL or the exact tokens to build against. NEVER treat mcp-figma or sk-design-md-generator as a taste, visual-direction, or critique authority — the extraction measures real CSS, it does not judge design.

11. Pass the design reference manifest to the dispatched session — when dispatching design or UI work, inline a DESIGN_DISPATCH_MANIFEST v1 block in the prompt (the child cannot resolve skill paths, so the manifest travels in the payload, not by reference): styleReferenceExtracted true, the live source that was measured, the measured design tokens / type scale / components the child must build against, loadedFiles, and proofDemandBack. If the manifest cannot be assembled — no Style Reference extracted, or no live source to measure — ASK before launching the child rather than starting a silent design dispatch. The child returns the demanded proof; the parent reconciles it on the return path.

12. Single-dispatch discipline (operator-gated, session-scoped) — Default: launch ONE cli-* dispatch at a time across the cli-* family (cli-opencode, cli-claude-code). Wait for the dispatched agent's work to return, verify outputs exist, then SIGKILL the dispatcher process + any orphan children (pkill -9 -f "claude -p" for this skill, plus gtimeout / positional_scoring_fallback:app cleanup). Only launch the next dispatch (this skill OR a sibling) after the prior one is dead and RSS has dropped. Within a deep-flow session (deep-review / deep-research): the operator authorizes the whole multi-iteration session at start — iterations chain back-to-back with kill-between as the safety mechanism, NOT a per-iteration operator confirmation prompt. Exception (cross-skill parallel): when the operator explicitly authorizes N parallel dispatches, run N concurrently — but still SIGKILL each as its work returns.

13. Set AI_SESSION_CHILD=1 in the dispatched child's env when sessions may be launched through the per-session worktree wrapper (.opencode/bin/worktree-session.sh). A dispatched claude -p run is an orchestrated sub-session, not a new top-level session, so it must SHARE the parent's worktree rather than allocate its own. The wrapper checks AI_SESSION_CHILD (plus a git --git-common-dir structural backstop) and exec's in place when set. Pattern: AI_SESSION_CHILD=1 claude -p .... Harmless when the wrapper is not in use. See .opencode/bin/README.md → "Worktree session isolation". Prepend SYSTEM_SPEC_GATE_ENFORCE=0 next to it so a dispatched child never inherits an enforced spec-gate from the parent shell (belt-and-suspenders alongside the wrapper's own neutralization and the core's complete AI_SESSION_CHILD classify/enforce no-op): SYSTEM_SPEC_GATE_ENFORCE=0 AI_SESSION_CHILD=1 claude -p ....

14. Agent-persona injection (attach identity, not just the task). Every dispatch composes {resolved agent persona + task prompt} — never a bare task. Resolve the persona from the ACTIVE runtime's agent directory per AGENTS.md §7 (.claude/agents/<name>.md, .opencode/agents/<name>.md, etc. — never hardcode one runtime), and map each subtask to the RIGHT agent (code→code, review→review, design→design, research→deep-research, docs→markdown), not one default. Claude Code has a native persona flag: pass claude -p --agent <name> — the CLI resolves .claude/agents/<name>.md on the print path, which satisfies the rule. On a bare claude -p that omits --agent, INLINE the persona block into the payload using the same in-payload pattern as the DESIGN_DISPATCH_MANIFEST (Rule 11) — the child cannot resolve agent paths by reference. A persona-less dispatch runs the leaf as a generic assistant, silently dropping the agent's tool-scope, verification gates, and output contract. Canonical contract: ../../sk-prompt/sk-prompt-models/assets/cli-prompt-quality-card.md "Persona Injection"; native precedent: orchestrate.md "Agent Loading Protocol". Rare exceptions (native surface used, focused summary for a small-context model, pure-mechanical command) are declared at the dispatch site.

⛔ NEVER

1. Use --permission-mode bypassPermissions without explicit user approval (auto-approves all writes/tool calls).
2. Trust output blindly for security-sensitive code (review for XSS, injection, hardcoded secrets, eval), or send sensitive data (API keys, passwords, credentials) in prompts — Claude Code transmits to Anthropic's API.
3. Hammer the API with rapid sequential calls — respect rate limits; use --max-budget-usd for cost control.
4. Use Claude Code for tasks where context is already loaded — direct action by the calling AI is faster.

⚠️ ESCALATE IF

1. Claude Code CLI is not installed and user has not acknowledged (provide npm install -g @anthropic-ai/claude-code).
2. Rate limits or budget caps are persistently hit (suggest increasing --max-budget-usd or checking quota).
3. Output conflicts with existing code patterns (present both perspectives; user decides).
4. Task requires --permission-mode bypassPermissions (describe risks; get explicit user approval).

Memory Handback Protocol

When the calling AI needs to preserve session context from a Claude Code CLI delegation, run the canonical 7-step procedure (extract MEMORY_HANDBACK section → build structured JSON → scrub secrets → invoke generate-context.js via --stdin/--json/temp-file → memory_index_scan). Full procedure and caveats: system-spec-kit/references/cli/memory-handback.md.

Claude-Code-specific Memory Epilogue template: see assets/prompt-templates.md §11.

Example invocation:

printf '%s' "$JSON_PAYLOAD" | node .opencode/skills/system-spec-kit/scripts/dist/memory/generate-context.js --stdin [spec-folder]

5. REFERENCES

Core References

• cli-reference.md - Complete CLI flags, commands, models, authentication, and configuration
• integration-patterns.md - Cross-AI orchestration patterns (reversed: external AI conducts, Claude Code executes)
• claude-tools.md - Unique capabilities and comparison with OpenCode
• agent-delegation.md - 9 agent roster, routing table, and invocation patterns

Templates and Assets

• prompt-templates.md - Copy-paste ready prompt templates for common tasks

Shared (cli-* family)

• shared-smart-router.md - Helper-function bodies for the smart router.
• memory-handback.md - Canonical 7-step Memory Handback procedure.

External

• Claude Code GitHub - Official repository
• Claude - Claude subscription OAuth account (auth for cli-claude-code)
• Claude Code Documentation - Official docs

Reference Loading Notes

• Load only references needed for current intent.
• Smart Routing (Section 2) is the single routing authority.
• cli-reference.md is ALWAYS loaded as baseline.

6. SUCCESS CRITERIA

Task Completion

• Claude Code CLI invoked with correct flags, model, and permission mode.
• Output captured, validated, and integrated appropriately.
• No security vulnerabilities introduced from generated code.
• Rate limits and budget caps handled gracefully.
• Appropriate agent routed for specialized tasks.
• Permission mode matched to task type (plan for review, default for generation).

Skill Quality

• All 8 sections present with proper anchor comments.
• Smart routing covers all intent signals with UNKNOWN_FALLBACK.
• Reference files provide deep-dive content without duplication.

7. INTEGRATION POINTS

Framework Integration

This skill operates within the behavioral framework defined in AGENTS.md.

Key integrations:

• Gate 2: Skill routing via skill_advisor.py
• Tool Routing: Per AGENTS.md Section 6 decision tree
• Memory: Context preserved via Spec Kit Memory MCP

Tool roles: Bash dispatches the CLI; Read/Glob/Grep validate output.

8. REFERENCES AND RELATED RESOURCES

The router discovers reference, asset, and script docs dynamically. Start with references/cli-reference.md, references/integration-patterns.md, assets/prompt-quality-card.md, assets/prompt-templates.md, references/agent-delegation.md, references/claude-tools.md, then load task-specific resources from references/, templates from assets/, and automation from scripts/ when present.

Related skills: cli-opencode for sandboxed OpenAI perspective, cli-opencode for full OpenCode runtime dispatch, sk-code for code-quality contracts, mcp-code-mode for external MCP work, and system-spec-kit for packet handback.