Deterministic structure. Minimal diffs.

Markdown edits an AI agent can trust.

Incise lets agents update Markdown tables, lists, sections, and YAML frontmatter by meaning—not fragile line numbers—while preserving unrelated document bytes.

Intent in. Exact edit out.

incise table-update-cell vault/Projects.md \
  --table Projects \
  --where Note=Launch \
  --column Status \
  --value done

The row is selected by content. Incise adjusts table padding when needed and leaves the rest of the file alone.

Semantic addresses

Target headings, cell values, item text, keys, and ordinals instead of stale line numbers.

Byte-preserving

Change the smallest structural range possible. Unrelated prose, formatting, and line endings survive.

Loud ambiguity

Unsafe or ambiguous operations refuse with an actionable remedy instead of choosing a plausible target.

Made for agent-managed Markdown

Tables and task lists

Update cells, add or remove rows and items, toggle checkboxes, and preserve formatting conventions.

YAML frontmatter

Set and delete nested keys while keeping comments, key order, scalar spelling, and document content intact.

Measured, not merely claimed

On the recorded small-model benchmark, structured Incise operations improved accuracy over direct patching for tables, lists, and sections. The repository publishes the tasks, raw calls, graders, caveats, and full findings.