Back to skill

Security audit

premortem

Security checks for vulnerabilities and agentic risk

Overview

This skill guides a structured planning exercise, with no evidence of hidden execution, sensitive access, or data collection.

Use it as a planning aid and share only project details you intend your agent to process. Review proposed plan changes and calendar entries before applying them.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
80% confidence
Finding

Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Content

Scanner excerpt · SKILL.md (reported line 49)May include surrounding context.

md
**Step 3 — Private silent generation:** each participant writes privately. No discussion. This step is non-negotiable — group discussion first produces conformity, not coverage.

**Step 4 — Consolidate:** round-robin read-out; facilitator captures all without judgment; group into clusters; rate each by probability / severity / detectability.

**Step 5 — Mitigate:** for each high-probability or high-severity cluster: mitigation action · owner · monitoring trigger · escalation threshold.

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
80% confidence
Finding

Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Content

Scanner excerpt · examples/backcasting-ai-agent-startup-failure-2024-2026.md (reported line 31)May include surrounding context.

md
### Step 4 — Consolidate

Round-robin read-out, captured without judgment, then clustered. Back-casting from the "we failed" endpoint produced causal chains, not just risk labels. The clusters that recurred across writers — the ones tied to durable, well-documented features of the 2024–2026 AI landscape rather than to this one team — were:

1. **Commoditized by the next release.** A subsequent foundation-model version, or a first-party feature shipped by the model provider or a platform, absorbed most of what the product did. Because the moat was an orchestration-and-prompt layer rather than proprietary data, deep workflow integration, or distribution, the differentiation evaporated the moment the base model got better and cheaper. This is the widely-discussed "is it a feature or a company?" problem for application-layer AI. Back-cast chain: *model upgrade → our wrapper's marginal value → drops toward zero → customers see no reason to keep paying → churn spikes → growth story dies.*