LLM Wiki — Compile Skill (wiki_compile)

Resolving script paths (read first): Commands below invoke scripts as <BIN>/X.py (and a few as <SKILLS>/...). Resolve these to absolute paths once before running anything:

>

- <SKILL_DIR> = the directory this SKILL.md lives in. - <SKILLS> = the skills/ folder containing this skill = <SKILL_DIR>/.. - <BIN> = the bin/ folder beside it = <SKILL_DIR>/../../bin

>

Do not hardcode a fixed prefix like .agents/bin or ../bin: shell relative paths resolve against the current working directory (usually the topic root), not this skill's location. Once resolved, <BIN> is typically .agents/bin when invoked from the hub root, or .claude/bin from inside a topic directory.

This skill handles the AI-driven "compilation" of high-entropy raw source texts into beautiful, structured, and interconnected literature cards and concept sheets.

Tooling (framework-agnostic): This skill is written tool-agnostic. Map each capability to your own agent's tool — read-file (Read in Claude Code, view_file in Antigravity), sub-agent / parallel task (Task/Agent in Claude Code, invoke_subagent in Antigravity), web-search (WebSearch in Claude Code, search_web), shell (Bash/PowerShell). Use the closest equivalent your framework provides; if a parallel sub-agent tool is unavailable, perform the steps sequentially yourself.

When the user asks to compile the wiki or process raw files, follow this parallelized workflow:

Phase 1: Preparation

1. Detect Uncompiled Sources: Run the helper script to detect uncompiled files programmatically:

python <BIN>/detect_uncompiled.py --topic-dir \"<TOPIC_DIR>\" This script output will list the relative paths of uncompiled source files.

Phase 2: Parallel Compilation (Subagents)

1. Invoke Subagents: Use your agent's sub-agent / parallel-task tool (spawning a fresh self-instance per task) to process one uncompiled raw file per sub-agent, concurrently. If your framework has no sub-agent tool, process the files sequentially yourself.
2. Subagent Task Instructions: Instruct each subagent to process its specifically assigned raw file independently:

• Strict Template Adherence: Read the raw document and apply the template <SKILL_DIR>/templates/paper_template.md.
• Ensure High-Density Output:
• No vague stubs: The compiled markdown must have rich sections covering math formulas (LaTeX), experimental metrics, contributions, and critical limits. If critical information is completely missing from the raw text, you MUST ONLY use the exact string [STUB: Awaiting synthesis] as the placeholder. Do not generate other variations like "No explicit definition".
• Bidirectional Linking: Apply standard Obsidian double bracket linking: [[Concept]] or [[Concept|Alias]]. Do NOT use standard markdown relative links for these.
• Concept Extraction (Thread-Safe): To extract novel concepts, do NOT write directly to wiki/concepts/. Instead, use the provided script to safely append your perspective. For each extracted concept, run:

python <BIN>/add_concept.py --name "Concept Name" --source "Source Paper Name" --content "Your detailed definition and perspective on this concept based on the paper you just read."

• Write Compiled Files: Save output ONLY under wiki/references/.
• Report Status: The subagent must report back to the main agent once the reference file is written and all concepts are extracted via the script.

Phase 3: Coordination and Cleanup (Main Agent)

1. Wait for Completion: The main agent must wait until all spawned subagents have successfully reported completion. If any subagent fails or times out, log the failure and continue.
2. Rebuild Navigation Indexes: Run the index builder script to deterministically rebuild all _index.md files:

python <BIN>/index_builder.py "<TOPIC_DIR>"

1. Run Final Linting: Run python <BIN>/llm-wiki.py lint --fix "<TOPIC_DIR>" on the full topic workspace as a final structural pass.
2. Log Activity: Log the compile event in log.md, including the count of successful vs. failed compilations.