trailofbits/skills

Project overview

It addresses a recurring need in AI-assisted security work — reusable, specialized workflows for analysis, verification, triage, and controlled tool use — spanning smart-contract scanners, static-analysis toolkits, verification workflows, and repository triage in one cataloged collection.

Project type

Evaluation & Observability

Deployment

Refer to project documentation

License

CC-BY-SA-4.0

Best for

• Developers and security engineers who already use a compatible agent host (ClaudeCode or compatible Codex routes) and want cataloged, specialized security analysis, verification, and triage workflows to select from.
• Teams willing to inspect each plugin's instructions, tool dependencies, and permission boundaries and to validate findings with independent evidence before acting on them.

Key capabilities

• Claude plugin marketplace, compatible Codex loading, and local parent-directory registration are documented. Workspace import uses .claude-plugin/marketplace.json; agents/openai.yaml requires its own display_name and short_description, and failed imports need resync after published metadata repair. No plugin is installed by the catalog itself.
• building-secure-contracts lists scanners for 6 blockchains and 5 guideline assistants; entry-point-analyzer finds state-changing contract entry points. Actual supported versions, detection coverage, and deployment safety are not verified.
• Audit-context building, source-parse-coverage C/C++ review, Rust safe/unsafe/memory/concurrency/FFI/async/SARIF review, dimensional unit/formula analysis, and sharp-edges API/configuration checks are cataloged. Descriptions do not guarantee complete vulnerability coverage.
• Differential-review uses history; variant-analysis searches related bugs; fp-check requires gate reviews; insecure-defaults uses per-file refuting verification; and vulnerability-triage-brocards accepts, dismisses, or requests information via 7 principles. These are workflow descriptions, not independent adjudications.
• A CodeQL/Semgrep/SARIF toolkit; rule creation and language variants with test validation; dependency advisory/version/publisher/install-script checks; Burp project extraction; GitHub Actions agent security audits; and TestingHandbook fuzzer/sanitizer/coverage skills are cataloged. Tool databases, credentials, and optional external analysis need separate configuration.
• trailmark describes code graph analysis, bounded subagent context slicing, Mermaid diagrams, mutation triage, and protocol verification. This catalog does not authorize delegation or actual code execution.
• Constant-time and compiler-side-channel review, secret zeroization in C/C++/Rust, surviving mutation analysis, reproducible post-patch variants/regressions, property tests using Hypothesis/fast-check/proptest/Echidna/Medusa, spec-to-code checking, and Lean4/Mathlib proofs are listed. Formal soundness and full security coverage were not established.
• YARA authoring includes lint and atom analysis; the DWARF skill parses and searches DIEs and integrity; FirebaseAPK scans for misconfiguration. No malware execution, binary parsing, APK scan, or external target probing is performed by the catalog itself.

Limitations and risks

• Individual plugins, AGENTS.md files, external tools, linked proof and bug reports, and the host marketplace implementation were not inspected. No plugin install, workflow, model, delegation, scanner, branch cleanup, merge, comment, or command was executed. Catalog claims are not independent assurance of security or compatibility.
• Selected plugins may read sensitive code or artifacts, contact host models or authenticated tools, and perform repository maintenance. Review permissions and dependency/provider terms before actual use. CC BY-SA 4.0 applies to the stated marketplace work, not blanket rights to third-party scanners, services, or source targets.

Getting started

• Confirm your host's supported marketplace behavior and the trusted provenance of plugins, including workspace-specific interface metadata where needed.
• Choose only the required plugin entries and inspect their instructions, tool dependencies, and account/configuration boundaries before use; the README does not define a universal dependency or permission matrix, and some development skills alter fetch routes, merges, or deletion workflows.
• Run analysis only on authorized artifacts, validate findings with independent evidence, and review maintenance or plugin actions before any external or destructive change.

Evidence and sources

• README: A Claude Code plugin marketplace from Trail of Bits providing skills to enhance AI-assisted security analysis, testing, and development workflows. Codex can load this marketplace…
• README: ``` /plugin marketplace add trailofbits/skills ```
• README: ``` /plugin menu ```
• README: Codex supports Claude plugin marketplaces directly, so this repository does not need Codex-specific sidecar metadata. Install the marketplace with: ```sh codex plugin marketplace…
• README: Use this repository's `.claude-plugin/marketplace.json` for workspace imports. Skills that include `agents/openai.yaml` also need `interface.display_name` and `interface.short_des…

Source updated at: 2026-10-11T20:23:16Z

Open original GitHub project