trail-of-bits-re

trail-of-bits-re — Trail of Bits security/RE skill packs

trailofbits/skills is a Claude Code plugin marketplace of ~40 installable skill packs distilling Trail of Bits' audit expertise. They are prompt/skill packs — no runtime, no MCP server, no vector-DB — so they ride the native /plugin bus and add zero always-on cost until invoked. Licensed CC BY-SA 4.0 (attribution + share-alike; keep the attribution if you fork a rule/skill).

This skill centers the three the E.5 gem calls out, then maps the rest so you can cherry-pick the RE/security-relevant ones without importing all 40 (catalogue tax — see the mcp-fleet-curation discipline).

The three core packs (the RE suite)

• differential-review — security-focused analysis of a code change. It folds in git history to spot vulnerabilities introduced by a modification (the class a plain unit-test diff sails past). Reach for it on PR review, a suspicious commit, or "did this refactor open a hole?" — complements the security-review skill (whole- branch) by focusing on the delta + its history.
• variant-analysis — once you have one confirmed bug, find its siblings: the same vulnerability class elsewhere in the codebase (or across repos) by pattern. This is the "we found CVE-X, where else does this pattern live?" workflow. Pairs with semgrep-rule-creator / semgrep-rule-variant-creator when you want to freeze the pattern into a reusable rule, and with the qlty/semgrep MCP to run it at scale.
• yara-authoring — author high-quality YARA-X detection rules (YARA-X is the Rust successor to legacy YARA) with linting, atom analysis (so your rule's literal atoms are actually selective, not a slow catch-all), and best-practice guidance. The most comprehensive pack in the collection. Reach for it whenever you turn a malware/sample finding into a durable signature — complements the RE stack (ghidra-mcp/ida-pro-mcp/capa-triage) by encoding what you found into detection.

Install (selective — user/setup step)

Claude Code:

/plugin marketplace add trailofbits/skills
/plugin menu        # browse, then install ONLY the packs you need

Codex:

codex plugin marketplace add trailofbits/skills
codex plugin list
codex plugin add differential-review@trailofbits
codex plugin add variant-analysis@trailofbits
codex plugin add yara-authoring@trailofbits

Layout is plugins/<name>/ per pack. Do not bulk-install all 40 — pick the RE/security-relevant handful below; every enabled pack is context budget you pay each session it loads.

/plugin marketplace add / install is an explicit /plugin-driven setup action — outside this skill's file scope. This skill documents the choice; it does not flip marketplace config for you.

The rest of the marketplace — RE/security-relevant picks

Beyond the core three, these are the packs worth considering for this owner's RE / Battle-Cats / security work (full set is ~40):

Pack	Use it for
dwarf-expert	reading/reasoning over DWARF debug info in native binaries
firebase-apk-scanner	scan an APK for exposed Firebase/backend config (BC-adjacent)
constant-time-analysis	crypto side-channel / non-constant-time comparisons
static-analysis	drive static analyzers with expert framing
insecure-defaults	flag dangerous default configs
sharp-edges	language/API footguns that read as correct but aren't
semgrep-rule-creator / semgrep-rule-variant-creator	turn a finding into a reusable Semgrep rule / rule variants
entry-point-analyzer	map a program's real entry points / attack surface
burpsuite-project-parser	parse a Burp project file for web-app RE
supply-chain-risk-auditor	dependency / supply-chain risk pass
zeroize-audit	verify secrets are actually zeroized in memory
second-opinion	independent adversarial review of your own conclusion
mutation-testing / property-based-testing / fp-check	strengthen test signal / cut false positives

(Others in the marketplace — building-secure-contracts, rust-review, c-review, modern-python, agentic-actions-auditor, workflow/meta packs — are useful but outside the RE core; install on demand.)

How it rides the existing bus

• Native /plugin — no new runtime, no MCP, no daemon. Just skills Claude reads.
• Complements, does not duplicate — differential-review + security-review (delta vs branch); variant-analysis + semgrep/qlty MCP (reason vs run-at-scale); yara-authoring + the RE disassembler stack (encode a finding into detection); the offline hacktricks KB stays the first-stop reference lookup.
• Advisory, not a gate — these packs make Claude reason better; none of them block, nag, or auto-fail anything. That is by design and satisfies the shadow-first posture (there is no behavior gate here to shadow).

Anti-overfit / posture

• Skill packs, not infrastructure — they reintroduce no rejected component (no vector-DB, no second orchestration runtime, no trained critic, no guardrail framework).
• Selective install is the discipline — bulk-enabling all 40 re-inflates the per-turn tool/token budget the mcp-fleet-curation skill exists to defend. Enable the RE core (3) plus a task-specific pick or two; retire what you stop using.
• CC BY-SA 4.0 — if you copy/adapt a rule or skill body, keep the attribution and the share-alike terms.

Verify

# skill is well-formed / vault-admissible (frontmatter parses, no angle brackets in description)
python - <<'PY'
import re, pathlib
p = pathlib.Path.home()/".claude/skills/trail-of-bits-re/SKILL.md"
t = p.read_text(encoding="utf-8"); fm = t.split("---",2)[1]
assert "\nname: trail-of-bits-re" in fm and "\ndescription:" in fm, "frontmatter missing"
desc = re.search(r"\ndescription:\s*(.+)", fm).group(1)
assert "<" not in desc and ">" not in desc, "angle bracket in description breaks quick_validate"
print("SUCCESS trail-of-bits-re SKILL.md well-formed")
PY

# after install, confirm the packs are present:
#   /plugin menu   (Claude Code)  or  codex plugin list  (Codex)