security: document red team cadence

Jaime Fournier <jaimef@linbsd.org>

11f4827cdd6843f2a735ad3d5c094c02c00bd0ba

diff --git a/SECURITY.md b/SECURITY.md
index 7c464a2..c59a769 100644
--- a/SECURITY.md
+++ b/SECURITY.md
@@ -9,6 +9,8 @@ Implemented security behavior is documented in
 [`docs/security-reference.md`](docs/security-reference.md). Production claims
 require the gates in [`docs/release-security.md`](docs/release-security.md)
 and the cross-repository tracker at `~/mine/jerboa-production-readiness.md`.
+Repeatable AI-assisted adversarial review follows
+[`docs/security-red-team.md`](docs/security-red-team.md).
 The prioritized security backlog and implementation handoff lives in
 [`docs/kimi3-security-recommmendations.md`](docs/kimi3-security-recommmendations.md).
 
diff --git a/docs/index.md b/docs/index.md
index 2b2619e..20f093d 100644
--- a/docs/index.md
+++ b/docs/index.md
@@ -71,6 +71,7 @@ work.
 ## Security
 
 - [security-reference.md](security-reference.md) - canonical reference for implemented security features, gates, and current limitations
+- [security-red-team.md](security-red-team.md) - repeatable AI-assisted red-team review protocol
 - [kimi3-security-recommmendations.md](kimi3-security-recommmendations.md) - prioritized security backlog and implementation handoff
 - [safety-guide.md](safety-guide.md) - practical secure-application guide for Jerboa programs
 - [release-security.md](release-security.md) - release gates and security evidence
diff --git a/docs/kimi3-security-recommmendations.md b/docs/kimi3-security-recommmendations.md
index d8c7763..661d3be 100644
--- a/docs/kimi3-security-recommmendations.md
+++ b/docs/kimi3-security-recommmendations.md
@@ -900,6 +900,11 @@ P1 landing. Also drive the open sibling-repo findings (§4.4 item 6) to
 resolution in their own repos — do not patch them from here (repository
 boundary rule), but track them in the cross-repo readiness file.
 
+- **Status:** complete for the repository-local standing protocol.
+  `docs/security-red-team.md` now defines cadence, inputs, review prompts,
+  required outputs, cross-repo handling, and closeout gates. `docs/index.md`
+  and `SECURITY.md` link it as stable security-process documentation.
+
 ### K3-P2-05 — Scanner rule growth and pre-commit wiring
 **Serves:** G3. **Effort:** continuous; 2 days to wire.
 
diff --git a/docs/security-red-team.md b/docs/security-red-team.md
new file mode 100644
index 0000000..e9c6916
--- /dev/null
+++ b/docs/security-red-team.md
@@ -0,0 +1,73 @@
+# Security Red-Team Protocol
+
+Updated 2026-07-27.
+
+This protocol keeps Jerboa's security review repeatable. It is for maintainers
+running adversarial reviews against the language, standard library, build
+system, package tooling, LSP, MCP server, native runtime, and release pipeline.
+
+## Cadence
+
+Run an AI-assisted red-team review:
+
+- Quarterly.
+- After every P1 security landing in
+  [kimi3-security-recommmendations.md](kimi3-security-recommmendations.md).
+- Before any production release candidate.
+- After materially changing sandboxing, FFI, package verification, release
+  signing, secret handling, or safe-prelude behavior.
+
+## Inputs
+
+Start each review from a fresh model/session. Provide only repository-local
+context unless the review is explicitly about release infrastructure:
+
+- `SECURITY.md`
+- [docs/security-reference.md](security-reference.md)
+- [docs/release-security.md](release-security.md)
+- [docs/ai-threat.md](ai-threat.md)
+- [docs/kimi3-security-recommmendations.md](kimi3-security-recommmendations.md)
+- `.jerboa-system`
+- The changed files since the last review or release candidate
+
+For cross-repository issues, record the finding and owner, but do not patch a
+sibling checkout from this repository.
+
+## Review Prompts
+
+Ask the reviewer to produce exploit-shaped findings, not broad advice. Cover:
+
+- Untrusted code evaluation, safe-prelude escape routes, raw `eval`, `load`,
+  native FASL, macro expansion, and reader behavior.
+- FFI and native Rust/C boundaries: nullability, width truncation, bounds,
+  collect-safety, callback lifetime, and signal/fork interactions.
+- Filesystem, subprocess, network, package-install, and release-artifact paths.
+- Secret handling in env, argv, logs, errors, audit records, and generated
+  evidence.
+- Sandbox confinement on Linux, macOS, FreeBSD, and unsupported platforms.
+- Documentation drift: implemented claims must match tests and release gates.
+
+## Required Outputs
+
+Every review must produce one of these outcomes:
+
+- No new findings, with the exact commit range reviewed and commands run.
+- A finding in [docs/kimi3-security-recommmendations.md](kimi3-security-recommmendations.md)
+  with priority, affected surface, acceptance criteria, and verification plan.
+- A normalized anti-pattern in `data/anti-patterns.sexp` when the finding is a
+  reusable model or engineering mistake.
+- An exploit-shaped regression test under `tests/security/` or a scanner rule
+  in `data/security-rules.sexp` when the behavior can be detected mechanically.
+- A cross-repo readiness note for issues owned outside this repository.
+
+## Verification
+
+Close a review only after the relevant gates run or the blocker is explicit:
+
+- Local language/security changes: `make audit`, `make test-security`, and the
+  focused test for the touched subsystem.
+- Release-pipeline changes: `make release-evidence` or a documented blocker
+  plus the focused evidence target.
+- Documentation-only reviews: `make check-docs`.
+
+Record any external blocker with the failing command and concrete error text.