Writing YARA Rules from Reversed Code

When to Use

• You finished reversing a sample/family and want a detection that survives recompilation and cosmetic changes.
• You need byte-pattern signatures from decryptors, API-hash constants, or unique algorithms rather than fragile strings.
• You are converting RE notes into hunting/scanning rules for a corpus.

Do not use volatile artifacts (file paths, mutex names that change per build, packer stubs shared across unrelated families) as your primary anchor — they cause drift and false hits.

Prerequisites

• The yara engine (and ideally yara-python) for testing.
• A small corpus: target samples (true positives) and clean/unrelated files (false-positive control).

Workflow

Step 1: Choose stable anchors

Prefer, in order: a unique algorithm's opcode sequence (decryptor, hashing loop), embedded magic constants (API hashes, XOR keys, S-box), then distinctive strings only if intrinsic.

Step 2: Extract byte patterns with wildcards

Pull the relevant opcodes and wildcard volatile operands (addresses, immediates) so the rule survives relocation/recompilation:

$dec = { 8A 04 ?? 34 ?? 88 04 ?? 41 3B ?? 7? ?? }   ; xor-decrypt loop, regs/disp wildcarded

Step 3: Assemble the rule

Combine 2–3 independent anchors with a condition requiring enough of them, plus a cheap prefilter (file size, PE magic) to keep scanning fast:

python scripts/analyst.py scaffold --name family_xyz --hash 0xABCDEF12

Step 4: Validate against the corpus

Run the rule across true positives and the clean control set; require all TPs match and zero FPs on the control.

yara -r rules/family_xyz.yar ./corpus

Step 5: Tune and document

Adjust thresholds, add meta (author, date, reference, hash), and record which construct each string anchors so future analysts can maintain it.

Validation

• Rule matches all intended samples and produces zero hits on the clean control set.
• Anchors map to intrinsic code/constants, not build-specific noise.
• meta documents source samples and the reasoning for each pattern.

Pitfalls

• Wildcarding too little (brittle) or too much (false positives); wildcard only volatile bytes.
• Anchoring on shared library/packer code present in unrelated binaries.
• Skipping the clean-corpus test, shipping a noisy rule.

References

• See references/api-reference.md for the rule scaffolder.
• YARA documentation and style guide (linked in frontmatter).