grep

name	grep
description	Search named IDA entities by pattern. Use when asked to find functions, labels, types, or members by name, or to seed xref/decompiler workflows from a name lookup.
metadata	{"argument-hint":"[search-pattern]"}
allowed-tools	["Bash","Read","Glob","Grep"]

grep is IDASQL's entity-search surface. Use it to discover named functions, labels, segments, structs, enums, and members before pivoting into xrefs, decompiler, or type work.

---

Trigger Intents

Use this skill when user asks to:

• find functions, labels, types, or members by name
• search by prefix/substring like sub_, EH, Zw, CreateFile, or main
• page through search results quickly
• seed table-native xref/decompiler/type workflows from name discovery

Route to:

• xrefs after locating a candidate callee/import/function and needing callers/callees/references
• decompiler after choosing a candidate function to inspect semantically
• types when the hit is a struct/enum/member you need to inspect or edit

---

Do This First (Quick Start)

-- 1) Start with a structured search while you learn the result shape
SELECT name, kind, addr
FROM grep
WHERE pattern = 'main'
ORDER BY kind, name
LIMIT 20;

-- 2) Narrow immediately when the result set is noisy
SELECT name, ordinal, full_name
FROM grep
WHERE pattern = 'EH%' AND kind = 'struct'
ORDER BY name;

-- 3) Page with ordinary SQL
SELECT name, kind, addr
FROM grep
WHERE pattern = 'sub_%'
ORDER BY kind, name
LIMIT 10 OFFSET 10;

Interpretation guidance:

• grep is a table. Use normal SQL for filtering, sorting, joining, grouping, and paging.
• For downstream parsing, select rows directly instead of wrapping the results in JSON.

---

Result Shape

grep exposes named IDA entities as rows:

• name
• kind
• addr
• ordinal
• parent_name
• full_name

Common kind values:

• function
• label
• segment
• struct
• union
• enum
• member
• enum_member

---

Pattern Rules

• Matching is case-insensitive.
• Plain text becomes a contains-match.
• % matches any substring.
• _ matches a single character.
• * is accepted and normalized to %.
• Empty pattern returns no rows.
• This is not regex.
• This is unrelated to byte_search.

Examples:

-- Contains-match
SELECT name, kind
FROM grep
WHERE pattern = 'main'
LIMIT 20;

-- Prefix wildcard
SELECT name, kind, addr
FROM grep
WHERE pattern = 'sub_%'
ORDER BY name
LIMIT 20;

-- Shell-style star is accepted too
SELECT name, kind
FROM grep
WHERE pattern = 'Zw*'
LIMIT 20;

---

Common Workflows

Find candidate functions by name

SELECT name, addr
FROM grep
WHERE pattern = 'main%' AND kind = 'function'
ORDER BY name;

Resolve imported APIs

SELECT module, name, addr
FROM imports
WHERE name LIKE 'CreateFile%'
ORDER BY module, name;

Find types by convention

SELECT name, kind, ordinal, full_name
FROM grep
WHERE pattern = 'EH%' AND kind IN ('struct', 'enum')
ORDER BY kind, name;

Find members under a parent type

SELECT name, parent_name, ordinal
FROM grep
WHERE pattern = 'flag%' AND kind = 'member'
ORDER BY parent_name, name
LIMIT 30;

Join into richer function metadata

SELECT g.name, f.size, f.prototype
FROM grep g
JOIN funcs f ON f.addr = g.addr
WHERE g.pattern = 'sub_%' AND g.kind = 'function'
ORDER BY f.size DESC
LIMIT 20;

Pivot from discovery into xrefs

SELECT caller_name, printf('0x%X', caller_addr) AS from_addr
FROM callers
WHERE func_addr = (
    SELECT addr
    FROM imports
    WHERE name = 'CreateFileW'
    ORDER BY name
    LIMIT 1
);

---

Compare With Other Search Surfaces

• Use grep for named entities discovered by IDA.
• Use strings when you need literal string contents.
• Use byte_search when you need raw bytes or opcode patterns.
• Use xrefs after discovery when the real question is "who references this?"

---

Failure and Recovery

• Too many hits: add kind = ..., tighten the prefix, or switch from plain text to a more specific wildcard pattern.
• No hits for an expected symbol: broaden the pattern, try a contains search, or pivot to imports if the target may only exist as an imported API.
• Need to search for comments, pseudocode text, or string contents: grep is the wrong surface; pivot to strings, decompiler tables, or other domain tables.
• Need bytes/opcodes: use byte_search instead of grep.