rlmgrep-usage - SKILL.md Agent Skill

name: rlmgrep-usage description: "Default repo search policy: whenever you need to search or read through files or directories, start with rlmgrep. Use rg/grep first only for strict literal/regex exhaustiveness or very large raw scans. Includes scoping and high-signal rlmgrep flag patterns."

Rlmgrep First

Overview

Default stance: if you need to search or read through files in a repo, start with rlmgrep.

If the request is even remotely fuzzy (conceptual wording, behavior-oriented question, likely cross-file reasoning, or "where/how/why" language), use rlmgrep before rg/grep.

Use rg/grep first only when the user explicitly needs:

strict literal/regex determinism,
exhaustive raw enumeration,
or maximum speed on very large corpora.

Default Workflow

Start with rlmgrep directly against the target repo/path.
If scope is broad, narrow with --type or -g and rerun rlmgrep.
If the user then asks for strict proof/exhaustive literal coverage, verify with rg.

High-Signal Commands

# Conceptual/cross-file search (default)
rlmgrep -C 2 "where is retry/backoff behavior implemented?" .

# Constrain scope while staying semantic
rlmgrep "where are api keys parsed and validated?" --type py .
rlmgrep "how is auth failure handled?" -g "**/*.py" -g "**/*.md" .

# Narrative grounded answer
rlmgrep --answer "how does this subsystem work end-to-end?" .

# Structured machine-friendly output
rlmgrep --signature 'summary: str, findings: list[str]' "audit auth behavior" .
rlmgrep --signature-json 'summary: str, risks: list[Literal["low","medium","high"]]' "assess risk" .

Key flags:

--answer for synthesized explanation before grep-style matches.
--paths-from-stdin when feeding a file list from another tool.
--type and -g to keep token/cost footprint focused.
-y to skip file-count confirmation prompt.

rlmgrep-First, rg-Second Pattern

Use this by default when there is any ambiguity:

# 1) semantic first
rlmgrep --answer "where is token lifetime enforced and what defaults apply?" .

# 2) literal verification only if needed
rg -n "token|expiration|ttl" .

Notes

rlmgrep is model-driven; it can find relevant lines even when wording differs from the query.
Regex-style prompts are best effort; use rg for strict regex guarantees.
Non-text search is supported (PDFs, Office docs, optional image/audio conversion).
Hidden and ignore files are respected by default; use --hidden/--no-ignore as needed.