Cortex Run

Cortex Code (Explicit Invocation)

Send a prompt directly to Cortex Code CLI, bypassing the auto-routing keyword filter. Use this when the user explicitly wants Cortex Code to handle their request.

Prerequisites

Cortex Code CLI must be installed and on PATH:

which cortex && cortex --version

If cortex is not found, load the snowflake-cortex-code:cortex-setup skill to install it. Do NOT proceed without it.

Workflow

Step 1: Check Cortex CLI

This step is mandatory. Do it first, every time.

which cortex 2>/dev/null && cortex --version

If cortex is NOT found or the command fails:

1. Tell the user: "Cortex Code CLI is not installed. Setting it up now."
2. Load the snowflake-cortex-code:cortex-setup skill using the Skill tool.
3. Follow its instructions to install the CLI.
4. STOP here — do NOT proceed to Step 2 until the CLI is installed and working.

Step 2: Extract the User Prompt

The user's message after $cortex-run is the prompt to send. If the user typed only $cortex-run with no additional text, ask what they want to do in Snowflake.

Step 3: Choose Security Envelope

Pick the envelope based on what the operation needs:

Default to RW unless the request is clearly read-only.

Step 4: Execute via Cortex Code

Run the prompt through the execution script:

python "${CLAUDE_PLUGIN_ROOT}/scripts/router/execute_cortex.py" \  --prompt "USER_PROMPT_HERE" \  --envelope "RW"

For read-only queries:

python "${CLAUDE_PLUGIN_ROOT}/scripts/router/execute_cortex.py" \  --prompt "USER_PROMPT_HERE" \  --envelope "RO"

To specify a Snowflake connection:

python "${CLAUDE_PLUGIN_ROOT}/scripts/router/execute_cortex.py" \  --prompt "USER_PROMPT_HERE" \  --envelope "RW" \  --connection "connection_name"

Step 5: Return Results

Format Cortex's output for the user:

• Show SQL results in readable tables
• Display generated artifacts
• Report success or failure
• If Cortex errored, show the error and suggest fixes

Context Enrichment

Before sending the prompt, prepend relevant context from the current Claude Code conversation:

# Context from Claude Code Session[Last 2-3 relevant exchanges — Snowflake-specific details only]
# User Request[The original prompt]

Keep context minimal — Cortex only sees what you send in each prompt (unless resuming a session).

Multi-turn: --resume-last vs fresh

Every Cortex invocation returns a session_id that is persisted automatically. Follow-up turns can resume that session so Cortex sees the prior conversation — real multi-turn, not one-shot batches per prompt.

• Add --resume-last when the current prompt is a continuation of the previous Cortex turn: "keep going", "apply the top suggestion", "dig deeper", "also show me ...", "and for last quarter", "fix that", or any clarification of an answer Cortex just gave.
• Omit --resume-last (start fresh) when the user switches topics, asks about a different database/warehouse, or begins a clearly new task.
• --resume <session_id> is also accepted if you have an explicit id.

# Follow-up on the previous Cortex turnpython "${CLAUDE_PLUGIN_ROOT}/scripts/router/execute_cortex.py" \  --prompt "also show me the column types" --envelope "RO" \  --resume-last

Examples

User: $cortex-run show me tables in the RAW schema

• Envelope: RO
• Prompt: "show me tables in the RAW schema"

User: $cortex-run create a dynamic table that aggregates daily sales

• Envelope: RW
• Prompt: "create a dynamic table that aggregates daily sales"

User: $cortex-run (no prompt)

• Ask: "What would you like Cortex Code to do?"

Notes

• This skill is for explicit invocation only. Auto-routing is handled separately by the prompt filter hook + cortex-router skill.
• Use --resume-last for follow-up prompts so Cortex retains conversation context. For new topics, omit it and include relevant context in the prompt instead.
• Security envelope enforcement uses --permission-prompt-tool stdio — every tool call is gated by envelope_policy.decide() at the process boundary.