Reliable Markdown operations for shell scripts and AI agents.

Texio extracts and surgically edits Markdown by document structure. Use it when regular expressions are unsafe and rewriting the complete file would create unnecessary changes.

grepfinds text.sedchanges text. Texio understands Markdown.

Install from crates.io with a current stable Rust toolchain, or use the platform-specific binary instructions.

cargo install texio-cli --locked

texio --versionCreate a small document and inspect its structure:

printf '# Demo\n\n## Installation\nold command\n\n## Usage\nkeep this\n' > demo.md

texio headings demo.md --json

texio section demo.md InstallationPreview one section change, then apply the same change after checking the diff:

texio replace demo.md --section Installation --text 'cargo install texio-cli --locked' --dry-run

texio replace demo.md --section Installation --text 'cargo install texio-cli --locked' --write

texio section demo.md InstallationThe preview leaves the file unchanged; the write preserves the Usage section. Copy the agent policy for future edits. Missing or duplicate headings cause an error rather than selecting a guessed target.

The four-fixture benchmark measured 82.8% fewer context-proxy tokens versus an idealized whole-file rewrite. This is not a model-token or billing measurement. A source build can take longer than five minutes; use a binary archive for the fastest first edit.

Preview the proposed change:

texio replace README.md \

--section "Installation" \

--from installation.md \

--dry-runApply it by replacing --dry-run with the explicit --write flag. Texio

preserves content outside the selected section and refuses ambiguous heading

matches.

texio headings README.md --json[{"level":1,"title":"Texio"},{"level":2,"title":"Installation"}]Install the published crate with a current stable Rust toolchain:

cargo install texio-cli --locked

texio --versionFrom a checkout, use cargo install --path . --locked.

For Linux x86-64, Windows x86-64, Intel macOS, and Apple Silicon macOS, follow the binary installation and checksum instructions. The same page documents the organization Homebrew tap and platform limits.

Texio is an early preview. The initial contract focuses on section extraction, heading discovery, and surgical replacement. CommonMark and GitHub Flavored Markdown compatibility work is ongoing.

Current parsing is powered by pulldown-cmark and supports ATX and Setext

headings while ignoring heading-like text inside fenced code blocks. Replacement

is atomic and preserves the target file's permissions.

See Markdown support for dialect coverage, editing guarantees, and current limitations. See the CLI contract for stdin/stdout behavior, JSON schemas, safety modes, compatibility policy, and exit codes.

AI agents frequently rewrite entire Markdown files to change one section. That uses unnecessary context and can alter unrelated content. Texio provides a small, deterministic operation that is easier to review and automate.

- Try ten tested agent recipes.

- Copy the agent instructions.

- Install the agent skill and review its automatic-selection evidence.

- Read Stop letting agents rewrite your entire README.

- Inspect and reproduce the public benchmark.

Across four Markdown fixtures, Texio used 45 context-proxy tokens instead of 261 for an idealized whole-file rewrite: 82.8% fewer tokens. Texio also passed 4/4 cases, compared with 3/4 for the whole-file baseline and 2/4 for the regex baseline.

The same result is provided below as stable, agent-readable data:

benchmark: markdown-agent-editing-v0.1.1

cases: 4

metric: unicode-word-and-punctuation-proxy

texio:

input_tokens: 32

output_tokens: 13

total_tokens: 45

passed: 4

whole_file:

input_tokens: 175

output_tokens: 86

total_tokens: 261

passed: 3

savings_vs_whole_file:

input_percent: 81.7

output_percent: 84.9

total_percent: 82.8

regex:

total_tokens: 45

passed: 2

source: benchmarks/results/v0.1.1.json

caveat: deterministic context proxy; not model tokenizer or billing tokensThese numbers measure the checked-in fixtures and CLI traffic, not every token an agent may consume while reasoning. See the methodology and raw result to reproduce or audit them.

MIT