Skip to main content

meta-security

Full security-audit orchestrator running /security-review, /cryptography-review, /threat-model, and /standards-review in canonical order per .github/skills/shared/orchestrator-contract.md. Each phase is a sub-agent (contract §6). Use before release or major changes; incremental skips only when explicitly requested for follow-ups.

Quellinformationen

Repository
cosmian/kms
Letzte Quellaktivität
15. Oktober 2026 um 14:44
Erkannte Sprache von SKILL.md
Englisch
Sterne
372
Forks
43

Installationsoptionen

Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.

Quelldateien prüfen

Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.

SKILL.md wird angezeigt

SKILL.md
Quellanweisungen · Schreibgeschützte Vorschau
name
meta-security
description
Full security-audit orchestrator running /security-review, /cryptography-review, /threat-model, and /standards-review in canonical order per .github/skills/shared/orchestrator-contract.md. Each phase is a sub-agent (contract §6). Use before release or major changes; incremental skips only when explicitly requested for follow-ups.
# Meta-Security — Comprehensive Security Audit Orchestrates all four security-focused skills in the correct order and produces a single unified go/no-go report. This is the most thorough security review available. > **When to use**: before a release (in addition to `/pre-release`), after a major > feature addition, when onboarding a new cloud provider integration, or when a > comprehensive security posture assessment is requested. --- ## Step 0 — Load Anti-Hallucination Discipline Read `.github/skills/shared/anti-hallucination.md` **before any analysis**. All rules in that file are mandatory for every sub-skill invoked below. Do not proceed until you have read it. ### Sub-Agent Routing Phases run as sub-agents (contract §6); the orchestrator contract in `.github/skills/shared/orchestrator-contract.md` sections 1 and 6 control: - **Phase ordering** (Section 1: security audit phases 1–4) - **Sub-agent routing** (Section 6: `slow` model for judgment/reasoning phases) - **Overlap deduplication** (Section 3: report overlaps once under earliest phase) - **Unified report schema** (Section 4: SUMMARY.md structure) - **Blocking criteria** (Section 5: go/no-go gates) This skill's manual steps below align with that contract. For orchestration rules and report format details, see the shared contract. ## Step 1 — Determine Scope Collect the union of branch changes, staged and unstaged changes, and untracked files. For example: ```bash git diff --name-only origin/develop...HEAD git diff --name-only HEAD git diff --name-only --cached git ls-files --others --exclude-standard ``` If `origin/develop` is unavailable, state that and use a known base or the supplied path; never treat an empty or failed diff as a clean scope. If a path was provided (e.g. `/meta-security crate/server/`), restrict all sub-skills to that path. Otherwise, use the full workspace. ### Full and Incremental Gates (per Orchestrator Contract Section 2) - **Full/comprehensive or release-oriented** requests run all four sub-skills. A supplied path narrows their scope; it does not itself authorize skipping a review. - **Incremental mode** applies relevance skips *only when the user explicitly requests* a follow-up. If scope or applicability is uncertain, run the review. - Record which areas changed and apply the conservative skip rules from the shared contract (Section 2). Record which areas changed: - [ ] `crate/crypto/` — crypto primitives - [ ] `crate/server/src/core/operations/` — KMIP operations - [ ] `crate/server/src/routes/` — HTTP routes - [ ] `crate/server/src/middlewares/` — auth/middleware - [ ] `crate/server/src/config/` — server config - [ ] `crate/server_database/` — database backends - [ ] `crate/hsm/` — HSM integrations - [ ] `crate/kmip/` — KMIP types - [ ] `crate/clients/` — CLI / WASM / client - [ ] `ui/` — Web UI - [ ] `.github/` — CI / scripts --- ## Step 2 — Security Review In full/comprehensive mode, always invoke `/security-review` on the scoped path. In an explicitly incremental review, apply the gate from Step 1 and record any skip. See `.github/skills/shared/orchestrator-contract.md` Section 2 (Skip Gates) for conservative skip rules. This phase covers (per `security-review` skill): - OWASP Top 10 / CWE Top 25 vulnerability families - Memory & type safety (FFI, `unsafe` blocks) - Deserialization & protocol parsing (TTLV, JSON, HTTP smuggling) - Race conditions / TOCTOU - Denial of service (ReDoS, resource exhaustion) - Side-channel attacks (timing, Marvin, Lucky13) - OAuth/OIDC & token-based auth attacks - HTTP-level & web security (CSRF, clickjacking, CORS) - Supply chain & dependency integrity - Security logging & monitoring - Business logic & KMIP-specific attacks - Injection flaws, secrets exposure, data handling - FIPS feature flag consistency - KMIP protocol authorization Collect all findings. Record the severity summary. If skipped in incremental mode, record the evidence and checked areas. ## Step 3 — Cryptographic Review In full/comprehensive mode, invoke `/cryptography-review` on the scoped path. In incremental mode, invoke it for crypto primitives/call sites, algorithm selection, key lifecycle/policy, provider initialization, or FIPS gates. Skip only after confirming none apply. See `.github/skills/shared/orchestrator-contract.md` Section 2 (Skip Gates) for conservative skip rules. This phase covers (per `cryptography-review` skill): - Algorithm inventory and FIPS/BSI/ANSSI compliance - Feature flag gating audit - Key size enforcement (multi-standard minimums) - OpenSSL provider initialization - Entropy / RNG audit - CBOM / SBOM currency - Key management lifecycle (SP 800-57) - Multi-standard compliance matrix - Academic research flags (known cryptanalytic attacks) Collect all findings. If skipped in incremental mode, record the checked areas and reason. ## Step 4 — Threat Model In full/comprehensive mode, invoke `/threat-model`. If a prior threat model exists, use incremental mode; otherwise use single analysis mode. For an explicitly incremental review, skip only after confirming no trust boundary or security-relevant data flow changed (authentication, routes, middleware, config/TLS, DB/HSM, crypto, external integrations). See `.github/skills/shared/orchestrator-contract.md` Section 2 (Skip Gates) for conservative skip rules. Collect all findings or record the evidence for an incremental skip. ## Step 5 — Standards Compliance Review In full/comprehensive mode, invoke `/standards-review` on the scoped path. In an explicitly incremental review, run it when protocol or standards-constrained behavior changed; otherwise record why it is not applicable. See `.github/skills/shared/orchestrator-contract.md` Section 2 (Skip Gates) for conservative skip rules. This phase covers (per `standards-review` skill): - KMIP 2.1 spec conformance (local HTML verification) - RFC conformance (URL-verified section citations) - FIPS / NIST SP conformance - BSI / ANSSI guideline conformance - Per-algorithm compliance checklist cross-reference Collect all findings. Record applicability and conformance gaps. --- ## Step 6 — Unified Report Produce a report using the unified schema from `.github/skills/shared/orchestrator-contract.md` Section 4 (Unified Report Schema). **Report structure** (adapted for security audit): ```markdown # Meta-Security — <Scope> Report **Generated**: <ISO 8601 timestamp> **Scope**: <full workspace | path | specific changed files> **Mode**: <full | incremental> ## Summary - **Total Findings**: N - **Critical**: N - **High**: N - **Medium**: N - **Low**: N - **Skipped Phases**: <list with evidence> ## Verdict **GO** / **NO-GO** (with reason) ## Findings (ordered by phase) ### Phase 1: Security Review [PASS | FAIL | SKIPPED] ... ### Phase 2: Cryptography Review [PASS | FAIL | SKIPPED] ... ### Phase 3: Threat Model [PASS | FAIL | SKIPPED] ... ### Phase 4: Standards Review [PASS | FAIL | SKIPPED] ... ``` ## Step 7 — Blocking Criteria (per Orchestrator Contract Section 5) Determine go/no-go using this order: ### BLOCKER (GO → NO-GO) - Any 🔴 **CRITICAL** finding from any phase - Any 🟠 **HIGH** finding from security-review or cryptography-review - Any 🔴 **Violation** from standards-review - Unmitigated **CRITICAL/HIGH** threats from threat-model - ⏭ **SKIPPED** phases are acceptable only in an explicitly requested incremental review with confirmed inapplicability; never skip in full/comprehensive or release mode ### WARNING (recorded but non-blocking) - **MEDIUM** findings; must be addressed before merge if count > 5 - **LOW** findings ## Step 8 — Output Rules - **Never** auto-apply fixes within this audit — present the unified report for human review (auto-fixing self-review is `/kms-pr-review`'s job) - **Always** attribute each finding to its source phase (per deduplication rules in shared contract Section 3) - **Always** deduplicate findings that appear in multiple phases (keep the most detailed version, note the overlap) - **Group** blocking findings at the top for immediate visibility - If all phases pass cleanly, say so clearly with a summary of what was scanned > **Note**: An incremental PASS is not a release go/no-go; use the full `/pre-release` gate before release.
Auf GitHub ansehen