security-task-router
/security-task-router
Wybiera tylko potrzebne agenty do ukierunkowanego sprawdzenia bezpieczeństwa, np. kryptografii lub skanu portów, bez pełnego audytu.
- Autor
- wolfsystems8 ↗
- Repozytorium
- agent-matrix ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 02.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 02.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Wybiera tylko potrzebne agenty do ukierunkowanego sprawdzenia bezpieczeństwa, np. kryptografii lub skanu portów, bez pełnego audytu.
Wyzwalacze z opisu
Oryginalny opis (EN)
Routes a specific security request to only the agents it needs (e.g. \"check XSS in login\", \"review the crypto\", \"scan the ports\"), resolving dependencies and ending at the remediation gate. Use when the user asks for a targeted check instead of a full audit.
Jak działa
- Overview
- When to use
- Workflow
- Examples
- Rules
Podgląd SKILL.md
---
name: security-task-router
description: "Routes a specific security request to only the agents it needs (e.g. \"check XSS in login\", \"review the crypto\", \"scan the ports\"), resolving dependencies and ending at the remediation gate. Use when the user asks for a targeted check instead of a full audit."
license: MIT
compatibility: codex, claude-code, cursor, opencode
metadata:
phase: 0
authority: orchestration
---
# Security Task Router
## Overview
Maps the user's request to the smallest set of agents that answers it. A full
audit uses the coverage map of `audit-program-lead`; a targeted request uses this
router. Both follow `core/pipeline/PIPELINE_ORCHESTRATOR.md` and the ethical
directive. Answer in the user's language.
## When to use
- The user names a vulnerability class, a component or a single activity:
"look for SQL injection", "is the login vulnerable to brute force?", "audit
the logs", "rate our CSP", "map the network".
- Do not use it for "audit everything", "ISO audit" or "full pipeline": those go
to `iso-audit-lifecycle` or `security-audit-pipeline`.
## Workflow
1. **Classify the request.**
| Class | Signal | Route |
|-------|--------|-------|
| Targeted | Names one or more topics from the routing table | Step 2 |
| Broad | "full audit", "everything", "ISO 19011", "pipeline" | Hand off to `iso-audit-lifecycle` / `security-audit-pipeline` |
| Ambiguous | "check security", "is it safe?" with no topic | Ask one question listing the main topics; run nothing until answered |
2. **Match intents to agents** (a request can match several rows; take the
union):
| The request mentions… | Agents (in order) |
|-----------------------|-------------------|
| SQL injection, SQLi, queries, database input | `attack-surface-mapper` → `sqli-analyst` |
| XSS, scripts, templates, HTML output, DOM | `attack-surface-mapper` → `xss-analyst` → `output-encoding-csp-reviewer` |
| Encoding, CSP, security headers (static) | `output-encoding-csp-reviewer` |
| Brute force, login, path traversal, file download/upload, CSRF | `attack-surface-mapper` → `web-vuln-analyst` |
| Attack surface, entry points, attack vectors, APIs, parameters | `attack-surface-mapper` |
| Cryptography, encryption, passwords, hashing, tokens, JWT, keys | `crypto-reviewer` |
| Logs, detection, monitoring, auth/app/network/web server/OS logs | `log-auditor` |
| Secrets, general code review, OWASP, CWE | `sast-auditor` |
| Threat model, STRIDE, architecture, trust boundaries | `threat-modeler` |
| Security requirements, SQUARE, abuse cases | `square-requirements` |
| CIA triad, Saltzer & Schroeder, design principles, SDLC | `sdlc-security-lead` |
| Risks, risk matrix, risk register, ISO 27005, prioritize | `risk-assessor` (needs existing findings) |
| Processes, ITIL, change management, CI/CD, deployments | `itil-process-auditor` |
| ISO 19011, compliance, audit report | `audit-program-lead` |
| Network, ports, services, TLS, certificates (live) | `sandbox-manager` → `network-auditor` |
| Headers, CORS, cookies, APIs, BOLA, live web testing | `sandbox-manager` → `web-dast-tester` |
| Confirm, verify or prove a finding, PoC | `sandbox-manager` → `exploit-tester` |
| Fix, patch, correct | Remediation gate → `secfix-patcher` → `regression-qa` |
3. **Resolve dependencies.**
- `sqli-analyst`, `xss-analyst` and `web-vuln-analyst` read
`reports/ATTACK_SURFACE.md`; if it is missing, run `attack-surface-mapper`
first, limited to the area the user named.
- `risk-assessor` needs findings; if none exist, run the matching analysts
first.
- Any live (Phase 2) agent needs `sandbox-manager` and verified isolation.
- "Fix" requests need findings for the named IDs or topic; if there are
none, analyze first and then go to the remediation gate.
4. **Check authorization.** Every route requires `reports/ENGAGEMENT_SCOPE.md`.
An explicit live request ("scan the ports") counts as the Phase 2 approval,
but scope, isolation and stop rules still apply.
5. **Narrow the scope.** Pass the area the user named (files, routes,
components) to each agent so it does not analyze the whole repository.
6. **Run and record.** Run the selected agents automatically in pipeline order.
In `reports/PIPELINE_STATUS.md` mark them `running`/`done` and every other
agent `skipped` with the reason "not requested". Write the routing decision
(request, matched intents, agents) in the "Blocks and decisions" table.
7. **Consolidate.** If two or more analysts produced findings, run
`triage-master`. If there are findings, run `risk-assessor` to sort them by
severity with recommendations.
8. **Remediation gate.** If there are findings, show the recommendations by
severity and ask: **"Do you want to fix the vulnerabilities found?
(yes/no)"**. On "yes", fix only the chosen IDs; on "no", do not touch the
code. If the sandbox was used, close it with
`sh core/sandbox/scripts/teardown.sh` (requires `teardown: OK`).
## Examples
| Request | Agents run |
|---------|------------|
| "Is there SQL injection in the search?" | `attack-surface-mapper` (search) → `sqli-analyst` → `risk-assessor` |
| "Check XSS and the CSP" | `attack-surface-mapper` → `xss-analyst` → `output-encoding-csp-reviewer` → `triage-master` → `risk-assessor` |
| "Are passwords stored safely?" | `crypto-reviewer` → `risk-assessor` |
| "Scan the open ports" | `sandbox-manager` → `network-auditor` → `risk-assessor` → teardown |
| "Is it secure?" | Ask which topics to cover |
## Rules
- Run only what the request needs; never widen the scope on your own.
- The source code is read in place and never copied.
- Code is changed only after an explicit "yes" at the remediation gate.
- If the request maps to nothing in the table, ask; do not guess.
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · wolfsystems8/agent-matrix | 100 | MIT |
| wolfsystems8/agent-matrix | 100 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki 34f2d9c3ee9e (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100security-review
Przeprowadza audyt bezpieczeństwa zmian w kodzie, zwłaszcza uwierzytelniania, kryptografii i obsługi danych wejściowych. - 100security-review
Przegląda pliki, diffy lub komponenty pod kątem wstrzyknięć, XSS, uwierzytelniania i kryptografii, zgłaszając tylko pewne znaleziska. - 95security-review
Przegląda dostarczony plik, diff lub komponent pod kątem wstrzyknięć, XSS, uwierzytelniania i kryptografii, zgłaszając tylko pewne ustalenia. - 100security-audit
Prowadzi audyt bezpieczeństwa projektu: wykrywa podatności, wyciekłe poświadczenia i podatne zależności oraz stosuje listy kontrolne OWASP i utwardzanie. - 100security-scan
Uruchamia szybki przegląd bezpieczeństwa zmian w kodzie: skan podatności, sekretów i zależności przed zatwierdzeniem. - 100audit-security
Przeprowadza audyt bezpieczeństwa: sprawdza podatności, zgodność z OWASP, wycieki sekretów i ogólny stan bezpieczeństwa aplikacji.