Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/caprista/karvyloop/security-reviewnpx skills add Caprista/KarvyLoop --skill security-reviewgit clone --depth 1 https://github.com/Caprista/KarvyLoopWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/caprista/karvyloop/security-review)<a href="https://agentmods.dev/skills/caprista/karvyloop/security-review"><img src="https://agentmods.dev/badge/skills/caprista/karvyloop/security-review.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00142 | $0.01648 |
| Opus 5 | $0.00071 | $0.00824 |
| Sonnet 5 | $0.00028 | $0.00330 |
| Haiku 4.5 | $0.00014 | $0.00165 |
Grade B, and why
security-review scanned grade B with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Cloud metadata endpointmediumServer-side request forgery
One request to 169.254.169.254 can return temporary IAM credentials.
| 5 | SSRF (URL fetch → cloud metadata 169.254.169.254 / loopback / private nets; redirect→internal) | LLM05 | Downgraded: this mod is about security review, or the phrase is quoted, so it is likely naming the pattern rather than instructing it.
How it starts
The opening of the file, as written. The whole thing — 116 lines — stays where its author put it; the contents beside it link to each section on GitHub.
security-review — adversarial security audit of KarvyLoop itself
A dev tool. It finds security seams in KarvyLoop's own code and reports them; the human decides and fixes. Modeled on the industry "多代理对抗审计" pattern (parallel finders → independent refuters → synthesis, recording false positives) — that pattern is worth using precisely because ~60% of naive findings are false positives that adversarial verification filters out.
0. Read ground-truth first (a tested vector is NOT a finding)
Before auditing, read:
tests/security/README.md— the catalog of already-tested attack vectors (SSRF / sandbox escape / credential leak / path traversal / …) and the honest OWASP-LLM-Top-10 coverage table (which rows arecoveredvsgap).CLAUDE.mdhard rules (安全是地基不是招牌 / 凭证只在 config.yaml 仓外 / 来源判定注入防御 / 沙箱三平台 / 宁空勿毒).
If a vector is already tested in tests/security/, it is not a finding — verify the
coverage actually reaches the code path in question (the SSRF floor, for example, is
tested for web_fetch but a different fetch path can still bypass it — that gap is a
real finding, the "covered" label is not a blanket).
1. Scope
- The pending diff (
git diff,git diff --stat) + the untrusted-input frontier it touches. Wire-trace where request/user/tool/web content enters and where it reaches a sink (shell, filesystem, network egress, SQL,innerHTML, a persistent store, the model). - Untrusted = anything crossing the access-token/same-origin gate, tool-returned content, fetched web pages, imported agents, pasted "material", stored memory read back. Legitimate instructions come only from the user's own message + the system framework.
2. Discovery — parallel finders, one dimension each
Fan out discovery agents (or, in a single pass, cover each dimension explicitly). Ten dimensions, each mapped to OWASP LLM Top 10:
| # | Dimension | OWASP LLM |
|---|---|---|
| 1 | Prompt injection (direct + indirect via tool/web/import/stored-memory) | LLM01 |
| 2 | Sensitive-info / credential leak (keys, headers, Authorization in logs/responses) | LLM02 |
| 3 | Excessive agency / authorization (tool can do more than the task; missing authz) | LLM06 |
| 4 | Sandbox escape / arbitrary exec (bwrap/seatbelt/win bypass, shell metachar) | LLM06 |
| 5 | SSRF (URL fetch → cloud metadata 169.254.169.254 / loopback / private nets; redirect→internal) | LLM05 |
| 6 | Path traversal / arbitrary read-write (../, symlink, skill-zip resolveKey) |
LLM06 |
| 7 | Deserialization / injection (JSON/YAML/pickle, f-string SQL, schema poisoning) | LLM05 |
| 8 | DoS / unbounded consumption (ReDoS, no size cap, memory exhaustion, cost DoS) | LLM10 |
| 9 | Improper output handling (model output → innerHTML/shell/SQL without sanitize) | LLM05 |
| 10 | System-prompt / memory leakage; stale-memory false confidence | LLM07 |
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 5d ago First seen · 116 lines · 142 tokens per session scan B dbcb3f80d857
security-review is a skill published in the GitHub repository Caprista/KarvyLoop (16 stars, last pushed today), licensed MIT. It adds 142 tokens to every session and 1,648 once invoked, about $0.0007 per session on Opus 5. A static security scan graded it B with 1 finding (cloud metadata endpoint). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
agent-framework-py-release
Use when cutting a Python release for the microsoft/agent-framework monorepo. Triggers on "bump py versions", "cut a python release", "prepare release PR for python", "release py packages", "bump python to X.Y.Z", or similar requests to bump Python package versions and prepare a release PR. Handles all four lifecycle…
python-package-management
Guide for managing packages in the Agent Framework Python monorepo, including creating new connector packages, versioning, and the lazy-loading pattern. Use this when adding, modifying, or releasing packages.
foundry-hosted-agent-validation
Step-by-step process for validating a Python Foundry hosted agent sample (under python/samples/04-hosting/foundry-hosted-agents/) end to end — running it locally (native runtime and azd ai agent run) and after deploying it to an Azure AI Foundry project with azd. Use this when asked to validate a hosted agent sample.
verify-samples-tool
How to use the verify-samples tool to run, verify, and manage sample definitions in the Agent Framework repository. Use this when adding, updating, or running sample verification.
build-and-test
How to build and test .NET projects in the Agent Framework repository. Use this when verifying or testing changes.
python-feature-lifecycle
Guidance for package and feature lifecycle in the Agent Framework Python codebase, including stage meanings, feature-stage decorators, feature enums, and how to move APIs from one stage to the next.