Share
## https://sploitus.com/exploit?id=47A03635-20F1-56F3-8AE9-04F92B01CCCF
# Security Skills

Security Skills is a Hermes Agent skill pack for defensive vulnerability research, CVE triage, exploit-intelligence workflows, code review, and security-tool-assisted analysis.

The project turns security work into repeatable, reviewable workflows. Instead of relying on ad hoc prompts, each `SKILL.md` defines when it should be loaded, what evidence it needs, which tools it expects, how it hands off to the next phase, and what output a user should receive.

## Who this is for

Use this repository if you want Hermes Agent to help with:

- triaging CVEs and choosing the right analysis path;
- researching advisories, patches, public exploit signals, and affected products;
- building reproducible local labs for authorized validation;
- reviewing security patches and hunting variants;
- writing and running Semgrep or CodeQL analysis;
- analyzing browser, kernel, WordPress, Android, Windows, Linux, web, logic, and memory-corruption vulnerabilities;
- standardizing security research notes, handoffs, and final reports.

These skills are intended for authorized defensive research, internal security review, and lab-based vulnerability validation.

## What is in this repository

### Core doctrine

- `ROUTING-DOCTRINE.md` β€” the source of truth for how the security skill family routes work, hands off between phases, and keeps skill behavior consistent.

### CVE workflow skills

The CVE workflow is designed as a deterministic pipeline:

```text
cve-identify-candidates
  -> cve-router-triage
    -> cve-research-analysis
      -> one target-specific branch skill
        -> cve-poc-validation
```

Primary CVE skills:

- `cve-identify-candidates` β€” turns a research intent into ranked CVE candidates.
- `cve-router-triage` β€” routes a named CVE to the right downstream skill.
- `cve-research-analysis` β€” builds the standardized CVE research dossier.
- `cve-poc-validation` β€” terminal-stage lab validation and reporting.

Target-specific branch skills:

- `cve-browser`
- `cve-kernel-linux`
- `cve-kernel-windows`
- `cve-logic-generic`
- `cve-memory-linux`
- `cve-memory-windows`
- `cve-mobile-android`
- `cve-web-generic`
- `cve-wordpress-workflow`

Exploit-development and Windows helper skills:

- `cve-memory-corruption-exploit-dev`
- `cve-windows-binary-diff`
- `cve-windows-debug-lab`
- `cve-windows-mitigation-bypass`

### Code-analysis and security-review skills

- `code-differential-review` β€” structured security review of patch commits and diffs.
- `code-variant-analysis` β€” hunt for similar bugs after one root cause is known.
- `code-semgrep-hunting` β€” write and run Semgrep rules for a known pattern.
- `code-codeql-analysis` β€” build and query CodeQL databases for deeper source-to-sink analysis.
- `design-sharp-edges` β€” identify dangerous defaults, footguns, and misuse-prone APIs.

### Tool installer

- `scripts/install-tools-ubuntu2404.sh` β€” idempotent Ubuntu 24.04 security-tool installer for local lab hosts.

The installer provisions common tools used by the skills, including Docker, GitHub CLI, Semgrep, EIP MCP, Ghidra MCP support, Android analysis tools, binary-analysis utilities, Burp Community, Metasploit, Wine, and supporting build/debug packages.

Run it from a clone of this repository:

```bash
sudo ./scripts/install-tools-ubuntu2404.sh
```

Optional toggles are documented at the top of the script, for example:

```bash
INSTALL_BURP=0 sudo -E ./scripts/install-tools-ubuntu2404.sh
INSTALL_DOCKER=0 INSTALL_BRAVE=0 sudo -E ./scripts/install-tools-ubuntu2404.sh
```

## How the skills work

Each skill is a self-contained `SKILL.md` with:

- frontmatter metadata for routing, tools, MCP dependencies, tags, and helper skills;
- clear β€œwhen to use” and β€œwhen not to use” guidance;
- prerequisite checks and required evidence;
- a step-by-step procedure;
- standardized handoff blocks for multi-stage work;
- final reporting expectations.

The important operating rule is: one skill owns the current phase. Helper skills can answer narrow questions, but they do not replace the current owner unless the handoff explicitly says so.

## Typical workflows

### Start from a research topic

Use `cve-identify-candidates` when you have a topic, vendor, product family, CWE, or time window but have not selected a CVE yet. It ranks candidates and pauses for user selection.

### Start from a known CVE

Use `cve-router-triage` first. It classifies the CVE and names the next skills to load.

### Review a security patch

Use `code-differential-review` when you have a patch commit, PR, advisory diff, or multi-file fix boundary to analyze.

### Hunt for variants

Use `code-variant-analysis` after a concrete root cause is known. If the pattern can be codified, hand off to `code-semgrep-hunting` or `code-codeql-analysis`.

### Validate a PoC

Use `cve-poc-validation` only after a branch skill has prepared a reachable target or bounded blackbox validation path. Acquisition-blocked and structurally blocked cases stop with a BLOCKED report instead of forcing PoC validation.

## Recommended lab layout

The CVE skills use the canonical lab layout:

```text
~/exploit-intel/
  labs/
    CVE-YYYY-NNNNN/
      INTEL.md
      report.md
      exploit.
```

`INTEL.md` is the resume point for future sessions. When returning to an existing lab, read it first and continue from the `NEXT_BRANCH` field instead of restarting the pipeline.

## Using these skills with Hermes Agent

Clone this repository:

```bash
git clone https://github.com/eip-public/security-skills.git
cd security-skills
```

Install the security toolchain on an Ubuntu 24.04 lab host if needed:

```bash
sudo ./scripts/install-tools-ubuntu2404.sh
```

Make the skill directories available to Hermes using your normal Hermes skill-loading workflow. For a local user install, copy or symlink the skill directories into your Hermes skills directory.

Example symlink approach:

```bash
mkdir -p ~/.hermes/skills/eip-cve
for skill in */SKILL.md; do
  dir="${skill%/SKILL.md}"
  ln -sfn "$PWD/$dir" "$HOME/.hermes/skills/eip-cve/$dir"
done
```

Then start a new Hermes session so the skill index is refreshed.

## Repository structure

```text
.
β”œβ”€β”€ ROUTING-DOCTRINE.md
β”œβ”€β”€ README.md
β”œβ”€β”€ code-*/SKILL.md
β”œβ”€β”€ cve-*/SKILL.md
β”œβ”€β”€ design-sharp-edges/SKILL.md
β”œβ”€β”€ security-skill-routing-graph.html
└── scripts/install-tools-ubuntu2404.sh
```

## Maintenance expectations

When adding or editing a skill:

- keep `ROUTING-DOCTRINE.md` current;
- keep frontmatter complete and standardized;
- define positive and negative scope;
- include explicit load triggers;
- use normalized handoff blocks;
- avoid stale tool names, paths, or retired workflow branches;
- verify with `git diff --check` before committing.

## Status

This repository is a working security skill pack. The current focus is standardization, reviewability, and reliable handoffs across the CVE and code-analysis workflow family.

## Contributing

See [`CONTRIBUTING.md`](CONTRIBUTING.md) for how to add a new skill, the frontmatter contract each skill must satisfy, and the routing-doctrine rules contributions are expected to follow.

## License

[MIT](LICENSE).