Back to skill

Security audit

ia-brainstorming

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed brainstorming workflow that reads project context and writes design notes, with no evidence of hidden execution, exfiltration, or unsafe persistence.

Before installing, understand that this skill is meant to read relevant project context and create design documentation, and it may commit that documentation when you have given authority. It is not an implementation skill and should not be treated as approval to change product code or publish work.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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 (3)

Hidden Instructions

High
Category
Prompt Injection
Confidence
70% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · SPEC.md (reported line 16)May include surrounding context.

md
Out of scope:
- Acting as the runtime instructions themselves (those live in `SKILL.md`).
- Trigger phrasings already covered by adjacent `ia-*` skills (`validate-plugin` flags >70% description overlap as DUPLICATE_TRIGGER).
- <!-- to fill in: domain-specific exclusions when the skill drifts -->

## Trigger Context

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
85% 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 · references/deep-interview.md (reported line 44)May include surrounding context.

md
Before exiting Phase 1, mentally combine what the user has stated so far. If stated-A + stated-B + agent-default-C produces a downstream effect the user is unlikely to have tracked (e.g., "if mute lives on the rule AND we don't warn on delete, rule-delete silently loses pause state"), fire one open probe per genuine combination. Phase 2.5's call-outs are a safety net for residuals, *not* a punt list for consequences you should have surfaced here.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · references/scope-synthesis.md (reported line 29)May include surrounding context.

md
**Path A vs Path B gate.** Routing depends on TWO signals: (1) did any *blocking* question fire before Phase 2.5? AND (2) what tier did Phase 0 classify? Blocking questions = scope disambiguation, dialogue probes, approach selection menus. Internal classification and pressure-tests do not count.

- **Path A**: Lightweight tier AND no blocking questions fired → announce-mode. Emit "What we're building" prose only (no other sections, no confirmation question), then proceed to Phase 3 doc-write in the same turn. Lightweight Path A docs are short; post-hoc revision is cheap.
- **Path B**: Standard/Deep tier OR any blocking question fired → full synthesis with confirmation gate. Two scenarios fire Path B: the user invested answer-time in dialogue, or pre-loaded substantive scope content. Either way, the substance earns a real checkpoint. The tier guard catches pre-loaded Deep brainstorms that would otherwise shortcut via the no-questions branch.

**Keep tests per section.** Each conditional section has its own keep test; failing items dissolve into the internal draft only.