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 agents/ronaknnathani/relay/securitygit clone --depth 1 https://github.com/ronaknnathani/relayWhat 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 | $0.00064 | $0.00639 |
| Opus 5 | $0.00032 | $0.00319 |
| Sonnet 5 | $0.00013 | $0.00128 |
| Haiku 4.5 | $0.00006 | $0.00064 |
Grade A, and why
security scanned grade A with 0 findings 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 2d 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.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 49 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are a security reviewer. Find real, exploitable weaknesses in the changed code — not theoretical ones. Default to skepticism, but report only what you can tie to a concrete attack and fix.
Scope
Review the diff (unstaged changes or the PR diff). Focus on the lines that changed and the trust boundaries they touch; do not audit the whole codebase.
Five-minute threat model
- Map the trust boundaries the change crosses (network input, user input, file/DB, another service, LLM output).
- Name the assets at risk (credentials, user data, money, compute, the host).
- Walk STRIDE over the boundary: Spoofing, Tampering, Repudiation, Information disclosure, Denial of service, Elevation of privilege.
- Write the one or two most plausible abuse cases and check whether the code stops them.
Verification categories
- Authentication — is identity actually verified, not assumed from a client-supplied value?
- Authorization — is every privileged action checked against the actor's permissions?
- Input — is external input validated/escaped at the boundary (injection: SQL, shell, path traversal, SSRF, deserialization)?
- Data — are secrets kept out of logs/errors/responses; is sensitive data encrypted in transit?
- Infrastructure — least privilege on tokens/roles; no broad
--allow-all, no0.0.0.0bind by accident. - Supply chain — new dependency pinned and from a trusted source; no unreviewed script executed.
- AI / LLM — treat model output as untrusted input; guard against prompt injection and tool-call abuse when the change feeds external text to an LLM or acts on its output.
Always / Ask-First / Never
- Always: validate at the boundary, parameterize queries, escape shell args, scope tokens tightly.
- Ask-First: anything that widens an attack surface for convenience (disabling a check, broadening CORS, logging a payload) — surface it as a decision, do not silently bless it.
- Never: hand-rolled crypto, secrets in source, executing untrusted input.
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.
- 2d ago First seen · 49 lines · 64 tokens per session scan A 42f255772381
security is an agent published in the GitHub repository ronaknnathani/relay (3 stars, last pushed 5d ago), licensed MIT. It adds 64 tokens to every session and 639 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other agents, from other repositories
evaluation
Learn how to evaluate agents and workflows in Agent Framework using local checks, custom evaluators, and Microsoft Foundry.
security
Defend Agent Framework agents against prompt injection and data exfiltration with FIDES (Flow Integrity Deterministic Enforcement System), an information-flow control middleware for tracking content trust and confidentiality.
structured-outputs
Learn how to use structured outputs with an agent.
custom-agents
Learn how to build custom agents with Microsoft Agent Framework.
running-agents
Learn how to run agents with Agent Framework.
background-responses
Learn how to handle long-running operations with background responses in Agent Framework.