Non-destructively annotate any Markdown text with bold, highlight, and code markup using an LLM, with an Anti-Tampering verbatim guard to prevent content loss.
Standalone Markdown annotation skill. Given any Markdown text, applies
**bold**, ==highlight==, and `code` markup to key concepts via LLM.
Designed to be reusable across any upstream skill that produces Markdown output (audio-transcriber, doc-parser, web-clipper, etc.).
markdown_text : str — raw Markdown string (any length)
subject : str — optional subject label for config profile selection
Annotated Markdown string. The Anti-Tampering guard ensures the LLM can never
delete content: if the output is shorter than verbatim_threshold × input length,
the original chunk is returned unchanged.
| Trigger | Behaviour |
|---|---|
LLM output shorter than verbatim_threshold × input | Original chunk restored |
| LLM exception on any chunk | Original chunk restored, processing continues |
config/config.yaml)| Key | Purpose |
|---|---|
model | Ollama model name |
chunk_size | Max chars per LLM call |
verbatim_threshold | Minimum output/input ratio (default 0.85) |
min_chunk_chars | Skip LLM for very short chunks |
| Caller | Phase | Input Source |
|---|---|---|
audio-transcriber | P4 Highlight | 03_merged/<subject>/<lecture>.md |
doc-parser | P2 Highlight | 01_Processed/<subject>/<pdf_id>/raw_extracted.md |