Interleave — Placeholder Filling Orchestrator

Fill <!-- LLM:NAME --> placeholders in a templated markdown document with subagent-generated content.

Trigger

Use when a markdown file contains <!-- LLM: placeholder markers — typically after running a template script from the interleave pattern.

Input

A path to a markdown file containing <!-- LLM:NAME --> placeholders.

If no path is provided, ask the user which file to process.

Steps

Step 1: Scan for Placeholders

Read the target file. Scan for all lines matching <!-- LLM: and extract the section names (the UPPER_SNAKE_CASE identifier after LLM:).

Report the count and names to the user:

Found N placeholders: NAME_1, NAME_2, ...

If no placeholders are found, report "No LLM placeholders found — document is fully rendered." and stop.

Step 2: Extract Context Per Placeholder

For each placeholder, extract:

• Task: the line starting with Task: (required)
• Format: the line starting with Format: (required)
• Context: everything between the Format line and END LLM:NAME -->

If Task or Format is missing, warn the user and skip that placeholder.

Step 3: Determine Model Tier

For each placeholder, select a model tier based on Task keywords:

Task Keywords	Model
"group", "cluster", "categorize", "organize"	sonnet
"summarize", "describe", "list", "extract", "synthesize"	haiku
"write", "compose", "draft", "envision"	opus
Default	haiku

Report the routing plan:

Routing:
  NAME_1 → sonnet (task: "Group these P2 items...")
  NAME_2 → haiku (task: "Write 2-3 sentence summaries...")

Step 4: Dispatch Subagents

For each placeholder, dispatch a subagent using the Task tool:

• subagent_type: general-purpose
• model: the tier from Step 3
• prompt: Include the Task instruction, Format guidance, and full Context block

All placeholders are independent — dispatch them in parallel using multiple Task tool calls in a single message.

Step 5: Replace Placeholders

For each completed subagent:

1. Take the subagent's output text
2. Find the placeholder block in the document (from <!-- LLM:NAME through END LLM:NAME -->)
3. Replace the entire block with the subagent's output

If a subagent failed or returned empty output, replace with:

<!-- WARNING: LLM:NAME unfilled — subagent returned no output -->

Step 6: Write and Report

Write the final document back to the same path.

Report:

Filled N/M placeholders in [path]
  NAME_1: filled (sonnet, ~X tokens)
  NAME_2: filled (haiku, ~Y tokens)
  NAME_3: WARNING — unfilled

Error Handling

• No file at path: Ask user for correct path
• No placeholders found: Report and stop (document is already complete)
• Malformed placeholder (missing closing tag): Warn and skip
• Subagent failure: Replace with warning marker, continue with other placeholders
• All subagents fail: Write document with warning markers, report errors