Ripgrep

ripgrep Text Search

Use ripgrep (rg) when the user needs fast text or regex content search from the command line. The useful outcome is a copy-pasteable command plus a clear statement of what it matches, what it skips, and why.

Triage

Start with the smallest tool that can answer the question.

• For trivial in-conversation lookups, use the harness's built-in ripgrep-backed search tool instead of shelling out to rg.
• Use rg for exact names, strings, regex content searches, file lists, counts, and pipeline-friendly output.
• Use ast-grep for syntax shape: descendants, ancestors, call forms, decorators, missing constructs, or nested contexts.
• Use language tooling for semantic facts such as type resolution, references, imports, or rename safety.

Before trusting local behavior, verify the binary: rg --version should print a ripgrep version plus enabled features such as +PCRE2. If it does not, run which rg — an alias or another tool named rg may shadow it.

Defaults Model

rg is a filter wrapped around a search. By default it:

• searches the current directory recursively (rg foo equals rg foo ./),
• respects .gitignore, .ignore, and .rgignore rules (precedence: .rgignore over .ignore over .gitignore),
• skips hidden files and directories,
• skips binary files (any file containing a NUL byte),
• does not follow symlinks (opt in with -L/--follow).

Escalation ladder: -u disables ignore-file handling, -uu also searches hidden files, -uuu also searches binary files — roughly grep -r with no smart filtering. Individual toggles: --no-ignore, --hidden, -a/--text.

When rg "can't find" a file, diagnose before changing the pattern:

1. rg --files | rg name — is the file in the searched set at all?
2. rg pattern --debug — which ignore rule or config excluded it? A * rule in the global gitignore is the most common cause.
3. rg pattern -uuu — does the match appear with all filtering off?

Workflow

1. Compose the simplest command that could work; prefer -F for literals.
2. Dry-run the file set with --files plus your -g/-t filters when the search scope matters.
3. Run on a narrow path first (rg pattern src/module), then widen.
4. For complex or shell-hostile patterns, write one pattern per line to a file and use -f patterns.txt (patterns are ORed). On Windows this is the preferred escape hatch from quoting problems; save the file as UTF-8 without BOM.
5. Report the command, what it matches, and what it may miss.

Filtering

• Paths: rg foo src tests README.md. Explicitly named files bypass ignore rules and hidden-file skipping.
• Globs: -g '*.toml' includes; -g '!*.toml' excludes — on the command line ! negates, the reverse of gitignore whitelist semantics. Later globs override earlier ones. Quote globs so the shell does not expand them. --iglob is the case-insensitive form. Glob patterns use / separators on every platform, including Windows.
• Types: -t py includes a type, -T js excludes one; --type-list shows the globs behind each type. --type-add 'web:*.{html,css,js}' defines a type for the current invocation only; persist it in the config file.
• Limits: --max-filesize 1M skips large files; --max-depth NUM caps traversal depth.

Regex Engines

• The default engine (Rust regex) guarantees worst-case linear time but has no lookaround and no backreferences — such patterns fail at compile time with a parse error.
• -P/--pcre2 switches to PCRE2, which supports lookaround and backrefs but can backtrack catastrophically. It requires a build with +PCRE2 in rg --version; otherwise rg reports PCRE2 is not available.
• --engine auto uses the default engine and falls back to PCRE2 only when the pattern needs it.
• -F treats the pattern as a literal string; -w requires word boundaries; -x requires the pattern to match the whole line; -e PAT or -- safely pass patterns that start with -.
• Case: -i insensitive, -s sensitive, -S smart-case (insensitive unless the pattern contains an uppercase letter).
• Multiline: -U lets a match span lines; add --multiline-dotall when . must match newlines. A bare \n in a pattern errors without -U.
• Unicode is on by default; disable it for a sub-pattern with (?-u:...).

Output & Pipelines

• Context: -A NUM, -B NUM, -C NUM.
• -o prints only the matched text; -r REPL rewrites the match in the output, supporting $1, $0, and named groups. See Hard Constraints.
• Machine-readable: --json emits JSON Lines with byte offsets and submatches; --vimgrep prints file:line:column:text with one match per line; --column adds column numbers.
• Lists and counts: -l prints files with matches, --files-without-match the inverse; -c counts matched lines per file — use --count-matches for total matches; --stats appends aggregate totals.
• Piping changes defaults: at a TTY rg shows colors, line numbers, and headings; when piped it drops them. Force with --color=always, -n, or --heading.
• Output order is non-deterministic because rg searches in parallel. --sort path gives stable order but disables parallelism; -j1 is a cheaper determinism knob.

Windows Quoting

• cmd.exe: only double quotes delimit strings; single quotes are literal characters, so rg '^foo' searches for a pattern containing quotes. Use rg "^foo". Escape a literal % as %%.
• PowerShell: prefer single quotes for regexes; $ inside double quotes triggers variable expansion, so rg "foo$" breaks while rg 'foo$' works.
• Git Bash / MSYS2: POSIX single quotes work, but a pattern or argument starting with / may be silently rewritten by path translation. Fix with a doubled slash (rg //foo) or MSYS_NO_PATHCONV=1 rg /foo.
• Escape hatch for any shell: put the pattern in a file and use -f file, saved as UTF-8 without BOM.
• UTF-16 files (common on Windows) are transcoded automatically via BOM sniffing; force an encoding with -E utf-16le when there is no BOM.

Hard Constraints

1. -r/--replace rewrites rg's OUTPUT only. ripgrep never modifies files and has no in-place edit flag. For real edits use the sanctioned pipeline rg foo -l -0 | xargs -0 sed -i 's/foo/bar/g' (BSD/macOS sed needs -i ''), or a purpose-built tool such as fastmod.
2. Zero matches usually means default filtering, not a bad regex. Walk the Defaults Model diagnosis (--files, --debug, -uuu) before rewriting the pattern.
3. Lookaround or backreferences on the default engine fail at compile time. Route to -P/--pcre2 after checking rg --version for +PCRE2, or restate the pattern without those features.

Debugging Checklist

Zero matches:

1. Verify the binary: rg --version and which rg.
2. rg --files | rg name — is the target file searched at all?
3. Add --debug — which ignore rule or config file excluded it?
4. Try -uuu — do matches appear with all filtering disabled?
5. Simplify: drop to a -F literal, then add regex features back.
6. On Windows, suspect shell quoting; move the pattern to -f file.
7. For non-UTF-8 files without a BOM, set -E explicitly.

Too many matches:

1. Scope with a path argument, -g, or -t.
2. Add -w for word boundaries or -x for whole-line matches.
3. Use -s when smart-case or -i over-matches.
4. Cap with -m NUM matching lines per file.

Output Contract

When answering an rg task, include:

1. Command: copy-pasteable, naming the target shell when quoting matters.
2. Why these flags: one line per non-obvious flag.
3. What it matches / misses: the filtering and engine boundaries in effect.
4. Caveats: ignore-file surprises, output ordering, encoding, or version gates (this skill documents ripgrep 15.1.0 behavior).

Reference

Load references/cli_reference.md when the task needs detailed flag semantics, the engine comparison, config-file format, preprocessors and compressed search, encodings, or JSON output details.