Text

AtlasClaw Workspace Output

When producing a user-facing .txt, .md, .markdown, or .html file:

1. Write only the final content under the current work_dir, preferably exports/<descriptive-name>.<ext> unless the user requested a safe relative filename.
2. Use UTF-8 text. For HTML, produce a complete standalone document when the user asks for an HTML file.
3. Expose the final file through the existing workspace download flow by passing its work_dir-relative path as download_paths.
4. Prefer writing a new output file instead of editing source files in place unless the user explicitly asks for an in-place edit.
5. Do not expose scripts, logs, caches, hidden files, intermediate files, or source documents as downloads.

AtlasClaw Runtime Commands

Prefer skill_read, skill_write, and skill_edit for simple file operations. If a transformation needs shell features such as pipes, redirects, globs, or compound commands, run it through bash -lc '<command>' with explicitly named work_dir-relative files, or use a python3 -c command that reads and writes explicit paths. Do not read or print .env, key, token, cookie, or password files unless the user explicitly requests it and the output is redacted.

Quick Reference

Task	Load
Text cleanup, extraction, Markdown, or HTML shaping	data.md
Translation/localization for text output	localization.md

---

Universal Text Rules

Encoding

• Always verify encoding first: file -bi document.txt
• Normalize line endings: tr -d '\r'
• Remove BOM if present: write a normalized copy instead of editing the source in place.

Whitespace

• Collapse multiple spaces: sed 's/[[:space:]]\+/ /g'
• Trim leading/trailing: sed 's/^[[:space:]]*//;s/[[:space:]]*$//'

Common Traps

• Smart quotes (\u201c, \u201d, \u2018, \u2019) break parsers -> normalize to ASCII quotes.
• Em/en dashes (\u2013, \u2014) break ASCII-only output -> normalize to -.
• Zero-width chars invisible but break comparisons → strip them
• String length ≠ byte length in UTF-8 ("café" = 4 chars, 5 bytes)

---

Format Detection

# Detect encoding
file -I document.txt

# Detect line endings
cat -A document.txt | head -1
# ^M at end = Windows (CRLF)
# No ^M = Unix (LF)

# Detect likely HTML
grep -qiE '<html|<!doctype html' document.html

---

Quick Transformations

Task	Command
Lowercase	tr '[:upper:]' '[:lower:]'
Remove punctuation	tr -d '[:punct:]'
Count words	wc -w
Count unique lines	sort -u | wc -l
Find duplicates	sort | uniq -d
Extract emails	grep -oE '[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}'
Extract URLs	`grep -oE 'https?://[^[:space:]<>"{}

---

Before Processing Checklist

• Encoding verified (UTF-8?)
• Line endings normalized
• Target format/style defined (TXT, Markdown, or HTML)
• Edge cases considered (empty, Unicode, special chars)