Stated plainly, up front, because it is the single most important thing to know before depending on this code:
moat has not been reviewed by an independent security auditor. It
implements well-understood patterns — token-bucket rate limiting, signed
double-submit CSRF tokens, os.Root-based path containment — carefully, and it
writes down its reasoning in doc/DESIGN.md so that the
reasoning can be checked rather than trusted. That is not the same as a security
guarantee, and it should not be read as one.
Treat this the way you should treat any security dependency: read the code, read the design record, and satisfy yourself that it does what you need. If you find that it does not, the process below is how to tell us.
The project is pre-1.0. Only the latest tagged minor version of each module receives security fixes.
| Module | Supported |
|---|---|
github.com/JonasBorgesLM/moat (core) |
latest v0.x tag |
github.com/JonasBorgesLM/moat/redisstore |
latest redisstore/v0.x tag |
examples/ |
not supported; illustrative code, not a released artifact |
Older v0.x tags are not patched. Because the API may still change between
minor versions, upgrading is the fix.
redisstore/v0.2.0 is retracted and does not compile; see
doc/DESIGN.md §6.16. go get will not select it, but a go.mod that pins it
explicitly still needs changing.
| Affects | Advisory | Status |
|---|---|---|
redisstore ≤ v0.2.1 |
GHSA-hfg8-hc9c-6c3h / CVE-2026-17106 (high) — moby/go-archive < 0.3.0, crafted tar writes outside the extraction directory |
Fixed in redisstore/v0.2.2. Not exploitable through this module; see below |
| core, all versions | none | — |
The go-archive advisory needs its scope stated precisely, because "high
severity in your dependency tree" and "your application is vulnerable" are not
the same claim and this library should not blur them.
go-archive arrives through testcontainers-go, which redisstore uses only in
its integration suite. Measured against the published v0.2.1 tag from a clean
directory outside the repository: the module appears in a consumer's go.sum
(both hashes) and in go list -m all, but none of its packages enter the build
(go list -deps finds nothing), and govulncheck on such a consumer reports No
vulnerabilities found. There is no path from redisstore's API to a tar
extraction. No binary built against v0.2.1 contains the vulnerable code.
What it does do is put a high finding into the dependency scanner of every
project that depends on redisstore/v0.2.1 — Dependabot, Snyk and
dependency-review-action read the module graph, not the reachability analysis —
and that finding cannot be cleared downstream, because the requirement is in this
project's go.mod. Handing that to downstream users is a real cost even with no
runtime impact.
v0.2.1 is therefore superseded but not retracted: it compiles, runs, and is
not exploitable, so a retraction would be the wrong instrument. Upgrade to
redisstore/v0.2.2 to clear the alert.
Dependabot alerts and automated security updates are enabled on the repository. They were not enabled before August 2026, which is how a high advisory reached a published tag; this one was reported within minutes of turning them on.
The toolchain a consumer builds with is part of this library's security posture, because the standard library is where most of the reachable vulnerability surface lives — the core module has no external dependencies at all.
| Module | go directive |
Why |
|---|---|---|
github.com/JonasBorgesLM/moat (core) |
1.24 | The floor set by os.Root, which pathguard is built on. No dependencies, and neither crypto/tls nor encoding/asn1 is reachable from it. |
github.com/JonasBorgesLM/moat/redisstore |
1.26.6 | Reaches crypto/tls via go-redis and encoding/asn1 via certificate parsing. GO-2026-6090 and GO-2026-5972 are fixed in 1.25.13 and 1.26.6. |
Two things about the redisstore floor are worth stating explicitly, because
both were wrong in redisstore/v0.2.1 and earlier:
- It is declared in the
godirective, nottoolchain. Atoolchainline in a dependency is ignored by whoever imports it, so a floor declared there protects this repository's CI and nobody else's build. That was the state up toredisstore/v0.2.1: the two vulnerabilities above were fixed for the maintainer and for no consumer. Thegodirective is honoured — under the defaultGOTOOLCHAIN=autoan older toolchain is upgraded automatically, and underGOTOOLCHAIN=localthe build fails with a clear message. - It is 1.26.6, not 1.25.13, though both carry the fixes. Go orders versions
across release lines, so a floor of
1.25.13is satisfied by1.26.5, which has neither fix.
If you build redisstore with an older toolchain by overriding this, both
vulnerable paths are reachable in your binary regardless of what this library
does.
Do not open a public issue, pull request, or discussion for a security problem. A public report is a disclosure: it tells everyone running the affected version, including people looking for something to exploit, before there is a fix to upgrade to.
Report privately through GitHub's private vulnerability reporting:
- Go to https://github.com/JonasBorgesLM/moat/security/advisories/new (or the repository's Security tab → Report a vulnerability).
- Fill in the advisory form. The report is visible only to the maintainers until it is published.
This channel requires a GitHub account and needs no key exchange or mailing list, which is why it is the primary one. If you cannot use it, open a public issue containing only a request for a private contact channel and no technical detail whatsoever, and a maintainer will follow up.
The more of this you can provide, the faster the fix:
- Affected module, package, and version or commit.
- The security property you believe is violated — for example "the rate limiter can be bypassed", "CSRF tokens validate across origins", "a path escapes the root".
- A minimal reproduction: ideally a failing Go test, otherwise concrete steps.
- Impact as you see it, and any conditions required (a specific
GOOS, a particular configuration, a non-default option). - Whether you intend to disclose publicly on your own schedule, and when.
You do not need a working exploit. A precise description of an unsound argument
in doc/DESIGN.md is a legitimate report; for a security library the reasoning
is part of the product.
| Stage | Target |
|---|---|
| Acknowledgement that the report was received | 3 business days |
| Initial assessment: confirmed, needs info, or not a vulnerability | 10 business days |
| Fix released, or a written plan with a date if the fix is complex | 90 days from confirmation |
These are targets from a small volunteer project, not a contractual SLA. If a deadline passes without a response, the report was probably lost — please ping the advisory thread.
We will keep you updated as the assessment progresses, credit you in the advisory unless you ask us not to, and let you review the advisory text before it is published.
We ask for the industry-standard 90 days from confirmation before public disclosure, and will usually publish sooner. If a fix is going to take longer than that, we will say so and explain why rather than let the clock run out silently. If a vulnerability is being actively exploited, we will move as fast as we can and disclose immediately after the fix ships.
When a fix ships we publish a GitHub Security Advisory with a CVE where one is
warranted, so that govulncheck and dependency scanners can see it. A security
fix that nobody's tooling can detect is only half a fix.
In scope — a failure of any security property this library claims:
- Rate limits that can be bypassed, including via key derivation, store
eviction, or a race in
Take. - CSRF tokens that validate when they should not: cross-origin, cross-user, forged, replayed after secret rotation, or accepted with a mismatched cookie.
- Timing side channels in token or secret comparison.
- Any input that causes
pathguardto read, write, or reveal the existence of a file outside its root, including via symlinks,.., absolute paths, Windows device names, or a TOCTOU race. - Response headers that are omitted, malformed, or weaker than documented.
validateguards that can be evaded — an unbounded body read pastMaxBodyBytes, a content-type check that a real parser disagrees with.- Panics reachable from untrusted input. The library must not panic in normal operation; a remotely triggerable panic is a denial of service.
- Unbounded memory growth driven by attacker-controlled input.
- Anything in
doc/DESIGN.mdthat is factually wrong about a threat.
Out of scope — real concerns, but not defects in this library:
- The absence of features it explicitly does not provide: it is not a WAF, an
IDS, a session manager, an authenticator, or an HTML sanitizer
(see
doc/DESIGN.md§1 and §6.4). - Limitations that are documented as such, unless you can show the
documentation is wrong. Notably:
MemoryStoreis per-process; same-origin XSS defeats CSRF protection entirely;pathguardconstrains paths and does not sandbox the process. - Vulnerabilities in
go-redis, or in any dependency of theexamples/module. Report those to their maintainers; if the flaw is in how we use them, that is in scope. - Insecure configuration chosen deliberately by an application — for example
WithFailureMode(FailOpen),WithoutCSP(), or aKeyFuncthat trustsX-Forwarded-Forunconditionally. If a default is unsafe, that is in scope and we want to hear it. - Findings from a scanner with no accompanying analysis of exploitability.
We will not pursue or support legal action against anyone who reports in good faith through the channel above, tests only against their own deployment, avoids privacy violations and service degradation, and gives us reasonable time to fix the issue before disclosing.
There is no bug bounty. This is an unfunded project; we can offer credit, a public thank-you, and a fast fix.