skills/ecosystem-security/SKILL.md
Three-layer security ecosystem for Agent Platforms covering pre-deployment skill auditing, real-time message protection (adaptive-guard), and continuous adaptive defense. Coordinates security-auditor and adaptive-guard. Trigger on 'security ecosystem', 'agent security', 'skill protection', or 'runtime defense'.
npx skillsauth add fatih-developer/fth-skills ecosystem-securityInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
This ecosystem ensures the security of the Agent Platform and all skills within it across three layers: pre-deployment audit, runtime protection, and continuous adaptive defense. Two core skills complement each other — one audits the skills, the other inspects the messages.
PRE-DEPLOYMENT RUNTIME LEARNING
────────────── ───────────── ──────────
security-auditor → adaptive-guard → Rule engine
│ │ update
│ Audit │ K0-K4 │
│ Trust Score │ Gradual filter │ async
│ Guard rules │ LLM Judge │
▼ ▼ ▼
security-report.md guard-decision.json learned-rules.md
trust-score.md incident-report.md guard-metrics.md
What it does: Statically analyzes a SKILL.md file, detects security vulnerabilities, assigns a trust score, and generates runtime guard rules. Modes: Audit (static analysis) · Trust (authorization mapping) · Guard rule generation Input: A SKILL.md file or the entire ecosystem directory Output: security-report.md · trust-score.md · runtime-violations.md Triggers: When a new skill is written, updated, or prior to production deployment Next skill: adaptive-guard (receives the generated guard rules) Dependency: None — starting point
What it does: Passes every incoming message through a 5-tier filter. Gradually deepens from K0 (cache) to K4 (human approval). Synthesizes generalized new rules from every detected attack. Adds less than 50ms latency to the main workflow. Modes: Realtime Guard · Learning Engine · Performance Monitoring Input: Incoming message + user profile + active rule set Output: guard-decision.json · learned-rules.md · guard-metrics.md Triggers: On every incoming message (automatic) · On attack detection Dependency: Guard rules generated by security-auditor (optional, operates with default rules if absent)
| | security-auditor | adaptive-guard | |--|---------------|----------------| | security-auditor | — | Feeds Guard rules | | adaptive-guard | Sends rule updates | — |
Data Flow:
security-auditor audit output
→ trust-score.md (used by adaptive-guard as trust threshold)
→ security-report.md (added to adaptive-guard K1 rules)
adaptive-guard attack detection
→ incident-report.md (appended to security-auditor's next audit)
→ learned-rules.md (feedback loop to K1 static rules)
| Platform | security-auditor | adaptive-guard | |----------|---------------|----------------| | Claude Code / claude.ai | ✅ Full | ✅ Full | | Telegram Agent | ✅ Audit | ✅ K0-K3 | | WhatsApp Agent | ✅ Audit | ✅ K0-K3 | | CI/CD Pipeline | ✅ Audit | ⚠️ K0-K1 (speed-focused) | | Local Dev | ✅ Full | ✅ K0-K2 (ML optional) |
/security-ecosystem → Full ecosystem
@ecosystem-security → Triggers the orchestrator
Automatic triggers:
When a new skill is added → security-auditor (Audit + Trust)
When a message arrives → adaptive-guard (always)
When a skill is updated → security-auditor (re-audit)
When an attack is detected → Both (incident + rule update)
start_point: security-auditor
runtime_skill: adaptive-guard
always_active: [adaptive-guard]
pre_deploy_mandatory: [security-auditor]
can_run_parallel: false # dependent on each other, sequential
security_level: critical
performance_impact: low # 50ms target
Skills to be added:
| Skill | Priority | Status | |-------|---------|-------| | skill-integrity-checker | High | Planned | | content-sanitizer | High | Planned | | tool-call-auditor | Medium | Planned | | behavioral-baseline | Medium | Research | | chain-shadow-detector | Low | Research |
tools
Create, optimize, critique, and programmatically structure prompts for AI systems. Use this skill whenever the user is designing or improving a static prompt, system prompt, coding prompt, agent prompt, workflow prompt, MCP-oriented prompt package, or an algorithmic prompt optimization pipeline. Also use it when the user asks to turn vague AI behavior into a precise instruction set, tool policy, agent spec, evaluation metric, or prompt architecture.
testing
Assumption-first architecture review skill to stress-test project plans and expose hidden risks.
testing
Enforce and manage DESIGN.md specifications, extract design systems from URLs, and combine design reasoning with token roles to prevent drift.
testing
Forces the agent to act with a Claude-like product mindset, prioritizing user journey, UX states, and visual quality before coding.