# Security Review

> Security-focused review of pending changes. Audit any change that could affect safety or correctness under abusive input or unchecked preconditions - oversized values, malformed structures, integer/length overflow, catastrophic backtracking, allocation pressure, other denial-of-service shapes, and any use of `unsafe` code or the `Unsafe` / `MemoryMarshal` / `Marshal` static helpers (which trade c…

- **Type:** Skill
- **Install:** `agentstack add skill-jeremykuhne-agent-skills-security-review`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [JeremyKuhne](https://agentstack.voostack.com/s/jeremykuhne)
- **Installs:** 0
- **Category:** [Developer Tools](https://agentstack.voostack.com/c/developer-tools)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [JeremyKuhne](https://github.com/JeremyKuhne)
- **Source:** https://github.com/JeremyKuhne/agent-skills/tree/main/skills/security-review

## Install

```sh
agentstack add skill-jeremykuhne-agent-skills-security-review
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Security review

If `overlay.md` exists beside this file, read it before acting; it contains
repository-specific bindings. This core remains usable without it.

Surface and pin behavior under **malformed input** and **caller-validated
APIs** (everything the C# compiler doesn't check for you).

## When to run

- The user asks for a security assessment / vulnerability review.
- A change adds or modifies a member that accepts caller-supplied data
  (function args, file contents, network bytes, env vars, anything
  upstream of those).
- A change introduces or modifies `unsafe` code, calls into
  `Unsafe.*` / `MemoryMarshal.*` / `Marshal.*`, `fixed`, raw pointers,
  unbounded `stackalloc`, or any BCL API whose XML doc says "unsafe"
  or "caller must". Applies even when inputs are fully internal -
  preconditions drift across refactors.
- A CVE in a comparable library prompts "do we have the same shape?".

## The discipline in one paragraph

Write tests that pin the **safe property** (terminates within a bound,
returns a defined error for malformed input, never reads/writes past the
buffer) - and make them pass against *current* code as a regression lock.
Never pin observable wrong output; that locks the bug in. Test the input
*shape*, not a specific exploit pattern. Boundary tests come in pairs (at
the limit, just over). When a finding needs a production change, surface
the options report and **end your turn** before patching. The full
principles and the High/Medium/Low rubric are in [principles.md](principles.md).

## Review procedure

1. **Inventory.** `git status --short`; tag each new/modified member that
   takes external input or uses a caller-validated API.
2. **Walk each tagged member** through the [checklist.md](checklist.md)
   categories. The largest category - `unsafe` / `Unsafe.*` /
   `MemoryMarshal.*` and other caller-validated APIs - has its own
   per-API table in [unsafe-apis.md](unsafe-apis.md).
3. **Place tests in `.Security.cs`** beside the production type.
4. **Add safe-property tests now; report findings that need a production
   change before patching.** See [reporting.md](reporting.md) for the
   options-report format and the don'ts.
5. **Run tests on every target framework** - timing bounds and allocation
   behavior differ across BCL versions.

When attention is limited, allocate it by tier (detail in
[checklist.md](checklist.md)):

- **Critical (never skip):** length/size bounds, algorithmic complexity,
  and the caller-validated-API audit.
- **Standard (every member):** integer overflow, malformed structure,
  argument validation.
- **Conditional (when the shape matches):** allocation DoS, in-band
  sentinels, path traversal.

## Related skills

Run alongside a broader pre-PR self-review before any publish. When you need
to *measure* a worst-case input rather than just bound it with a `Stopwatch`,
use the repository's performance-testing skill. (A consuming repository wires
the concrete cross-references in its overlay.)

## Sub-pages

- [principles.md](principles.md) - the five core principles and the
  High / Medium / Low severity rubric.
- [checklist.md](checklist.md) - the nine audit categories with the test
  shape each one demands.
- [unsafe-apis.md](unsafe-apis.md) - the per-API precondition / failure /
  test-shape table for `unsafe`, `Unsafe.*`, `MemoryMarshal.*`, `fixed`,
  `stackalloc`, and `[DllImport]`.
- [reporting.md](reporting.md) - the review procedure, the options-report
  format, and the don'ts that keep tests from pinning bugs.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [JeremyKuhne](https://github.com/JeremyKuhne)
- **Source:** [JeremyKuhne/agent-skills](https://github.com/JeremyKuhne/agent-skills)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-jeremykuhne-agent-skills-security-review
- Seller: https://agentstack.voostack.com/s/jeremykuhne
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.