Add stdin/file support for invoke prompt to support long payloads

[DennisTraub]

Summary

agentcore invoke only accepts prompts as a positional argument or --prompt flag. There's no way to pipe a prompt from stdin or read it from a file. For agents and scripts working with long prompts (multi-turn conversations, structured JSON payloads), this hits shell argument length limits and forces awkward workarounds.

Current behavior

# Works for short prompts
agentcore invoke "What is the weather?" --json

# Fails for long prompts - shell argument limit
agentcore invoke "$(cat long-prompt.txt)" --json

# No stdin support
echo "my prompt" | agentcore invoke --json
# prompt is empty - stdin is ignored

Suggested fix

Either or both:

1. --prompt-file <path> flag that reads the prompt from a file
2. Stdin support - when no prompt is provided and stdin is not a TTY, read from stdin

// Example: stdin support
if (!prompt && !process.stdin.isTTY) {
 const chunks: Buffer[] = [];
 for await (const chunk of process.stdin) chunks.push(chunk);
 prompt = Buffer.concat(chunks).toString('utf-8').trim();
}

Why this matters

AI agents composing multi-step workflows frequently need to pass structured or lengthy prompts to deployed agents. Shell argument limits (typically 128KB-2MB depending on OS) become a real constraint. Stdin/file support is the standard escape hatch and enables clean piping:

# Read prompt from file
agentcore invoke --prompt-file prompt.json --json

# Pipe from another command
jq -r '.response' result.json | agentcore invoke --json

# Agent chaining
agentcore invoke agent-a "summarize this" --json | \
 jq -r '.response' | \
 agentcore invoke agent-b --json

This follows conventions from gh (--body-file), aws (file://), and similar CLIs.