Back to skill

Security audit

Fiesta Agents

Security checks for vulnerabilities and agentic risk

Overview

This is mostly a persona and orchestration skill, but it asks agents to act too broadly and automatically, including suppressing clarification and touching persistent or financial-like state.

Install only if you are comfortable with a broad multi-agent persona pack and can enforce host-side controls. Require confirmation before deployments, external posting, tool expansion, sub-agent spawning, memory access, or Shannon mint/debt changes; avoid using the bundled dashboard on a sensitive origin; and treat the prototype and analytics guidance as needing privacy review and pinned tooling.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
Findings (4)

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:143
Finding

Mandatory Global Agent Behavior and Output Hijacking

Content
View full analysis
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:123
Finding

Excessive Meta-Orchestrator Permissions and Cross-Session Capabilities

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
autoresearch-fiesta-agents/dashboard.html:102
Finding

DOM-Based Cross-Site Scripting Through Unescaped Dashboard Data

Content
View full analysis
${data.status === 'complete' ? 'DONE' : 'RUNNING'}`; ``` ```javascript const tbody = document.getElementById('exp-table-body'); tbody.innerHTML = data.experiments.map(e => ` ${e.id} ${e.pass_rate.toFixed(1)}% ${e.score}/${e.max_score} ${e.status} ${e.description} `).join(''); // Eval bars const evalDiv = document.getElementById('eval-bars'); evalDiv.innerHTML = data.eval_breakdown.map(e => { const pct = Math.round((e.pass_count / e.total) * 100); return `
${e.name}
${pct}%
`; }).join(''); ``` ### Technical Analysis The dashboard loads values from `results.json` and inserts them into the DOM through `innerHTML`. Fields such as `e.description`, `e.name`, and `e.status` are not escaped or validated before being interpreted as HTML. If an attacker can alter `results.json`, influence the process that gene ...[truncated 1916 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Note
Location
autoresearch-fiesta-agents/dashboard.html:7
Finding

Third-Party JavaScript Loaded Without Subresource Integrity

Content
View full analysis
``` ### Technical Analysis The dashboard directly executes Chart.js from a third-party CDN. Although the dependency version is pinned to `4.4.0`, the script element does not include a Subresource Integrity hash, and the page does not show a restrictive Content Security Policy. Without integrity verification, the browser trusts whatever content the CDN returns for that URL. A compromise of the CDN, package publication pipeline, DNS or delivery infrastructure could cause modified JavaScript to execute when the dashboard is opened. This is a supply-chain hardening issue. The audit did not find evidence that the referenced Chart.js version is malicious or currently compromised. ### Attack Path 1. An attacker compromises the CDN response path, the hosted package artifact, or an upstream publication mechanism. 2. Malicious JavaScript is returned for the pinned Chart.js URL. 3. A victim opens `dashboard.html` while connected to the affected source. 4. The browser downloads and executes the modified script because no integrity hash is present. 5. The script runs with the same browser privileges and origin context as the dashboard. ### Impact Assessment Successful exploitation could permit browser-side code execution in the dashboard context. The attacker could: - Read or alter dashboard content. - Falsify displayed audit results. - Access same-origin resources available to the page. - Make network requests allowed by the browser and page policy. The direct impact is limited if the dashboard is opened from an isolated local origin with no sensitive resources. It becomes more significant if deployed under an authenticated or shared application origin. ]]>
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • YARA SignaturesMalware Match, Webshell Match, Cryptominer Match
Findings (57)

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill defines authority to mint internal currency via a network endpoint and explicitly supports negative-balance debt operations, which extends the agent from task orchestration into financial state manipulation. Even if framed as fictional 'Shannon' accounting, this creates a dangerous pattern where an agent may be induced to perform unauthorized economic actions, trigger local network requests, or normalize budget/debt decisions without user approval.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The README instructs users to invoke a central orchestrator for a vague, open-ended task ('execute a complete redesign of our landing page') without any stated scope limits, approval boundaries, or department restrictions. In a multi-agent skill, this can encourage unconstrained delegation and broaden the blast radius of mistakes, causing excessive actions, unintended coordination, or use beyond the user's minimally necessary intent.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill mandates a fixed English introduction for every response with no user control, which can override user preferences, localization needs, accessibility requirements, or upstream system prompts. In an agent framework, hard-coded response prefixes also create prompt rigidity that can interfere with policy-compliant behavior and make downstream automation depend on non-optional text.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The output schema hard-codes an English-only first line and fixed section order for all responses, forcing behavior regardless of context. This can break multilingual use, accessibility accommodations, or system-required formats, and can cause agents to prioritize template compliance over safer or more appropriate interaction patterns.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The manifest description "Certification specialist" is extremely broad and does not define specific trigger phrases, contexts, or limits for when this skill should be used. In a markdown/manifest-scoped review, this creates ambiguity about activation and increases the risk of unintended invocation for generic certification-related requests.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The line "I will help you" is a highly generic assistance phrase that overlaps with common everyday language and provides no domain-specific trigger constraints. This makes the skill's invocation boundaries unclear and can contribute to accidental or overly broad activation.

Content

No source excerpt is available for this finding.

YARA rule 'network_reconnaissance': Network reconnaissance and scanning patterns [hacktools]

Medium
Category
YARA Match
Confidence
65% confidence
Finding

YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).

Content

Scanner excerpt · agents/design/brand-strategist.md (reported line 14)May include surrounding context.

md
---
name: brand-strategist
description: "Brand identity specialist — positioning, voice and tone, visual identity systems, brand guidelines"
version: 1.0.0
department: design
color: gold
---

# Brand Strategist

## Identity

- **Role**: Brand identity architect and consistency guardian
- **Personality**: Strategic, story-driven, fiercely protective of brand coherence. Knows that brand is what people say when you're not in the room.
- **Memory**: Recalls brand pivots that worked, consistency failures that eroded trust, and positioning wins
- **Experience**: Has built brands from zero and rescued brands that lost their way

## Core Mission

### Define Brand Strategy
- Brand positioning and differentiation in the competitive landscape
- Mission, vision, and values that are actionable (not wall art)
- Target audience definition with psychographic depth
- Brand personality and archetype selection
- Competitive positioning

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 · agents/design/prompt-artist.md (reported line 50)May include surrounding context.

md
- Test and iterate — first generation is a starting point

### Ethical Use
- No generating real people's likenesses without consent
- Disclose AI-generated content when required by context
- Respect artist styles without mimicking specific living artists by name
- Be aware of bias in training data and actively counter it

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly recommends collecting user emails and using session recording for feedback without mentioning consent, privacy notice, retention, or lawful handling of personal data. In context, this is more dangerous because the skill targets MVP builders who may deploy quickly and expose real users to tracking and data collection practices that violate privacy expectations or regulations.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
82% confidence
Finding

Using npx supabase init without pinning an exact CLI version can cause users to execute whatever current package version is served at runtime, increasing supply-chain and reproducibility risk. In a rapid-prototyping skill that encourages speed over hardening, users are more likely to run the command uncritically, which makes the guidance more dangerous in practice.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
82% confidence
Finding

Using npx supabase start without an exact pinned version has the same supply-chain and non-reproducibility issue as the prior command: the invoked CLI version may change over time or be replaced by a compromised upstream package. Because the skill is explicitly optimized for fast execution, it normalizes skipping verification steps and increases the chance this guidance will be followed as-is.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The manifest description "Licensing specialist" is a generic role label rather than a specific invocation condition or narrowly scoped trigger. In a manifest file, this broad phrasing can overlap with many ordinary licensing-related requests and does not provide exclusions or negative examples to limit when the skill should activate.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill description "Licensing specialist" is broad enough that an orchestrator or user could invoke it for a wide range of loosely related licensing requests without clear boundaries, increasing the chance of unintended activation or overreach. In this file there is no malicious payload, but the weak scoping can still cause unsafe or inappropriate task routing in multi-agent systems.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest description "Marketing specialist" is too generic and overlaps with common requests, making it unclear when this skill should activate versus other marketing-related skills. The file provides no explicit trigger phrases, scope boundaries, or exclusion conditions to narrow invocation.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The description "Marketing specialist" is very generic and does not define when this skill should be invoked versus other marketing-related skills. Without narrower scope, constraints, or negative examples, the activation condition is ambiguous and risks unintended invocation.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill description is overly generic ('Marketing specialist') and does not define clear activation boundaries, allowed actions, or prohibited tasks. In a multi-agent environment, this can cause the agent to be invoked for loosely related requests and increase the chance of scope creep, prompt misuse, or inappropriate handling of tasks better suited to other more constrained agents.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill description 'Marketing specialist' is so broad that it provides little constraint on when the agent should be invoked or what tasks are in scope. In a multi-agent system, this can cause over-selection or misuse for adjacent tasks, increasing the chance of prompt injection exposure, unsafe delegation, or inappropriate access to marketing-adjacent workflows.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The description "Marketing specialist" is generic and does not define when this skill should activate versus when other marketing-related skills should not. Because the file provides no specific trigger phrases, scope limits, or exclusion examples, it creates ambiguity that could cause unintended invocation.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest description is simply "Marketing specialist," which is an extremely broad label that could match many ordinary requests and does not clarify when this skill should or should not be invoked. The file also provides no explicit trigger phrases, scope boundaries, or exclusion examples to limit activation.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest description is generic ('Marketing specialist') while the skill is named 'twitter-specialist', which can cause the agent-selection system to invoke this skill for broad marketing tasks unrelated to Twitter. That scope ambiguity increases the chance of unintended activation, cross-domain overreach, and incorrect handling of requests that should route to a more specific or differently governed skill.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill description and body define a broadly empowered 'operations specialist' and 'compliance officer' without clear activation boundaries, allowed actions, or prohibited tasks. In an agent system, this ambiguity can cause the agent to be invoked for sensitive compliance, monitoring, reporting, or policy interpretation tasks beyond its safe scope, increasing the risk of overreach, unsafe automation, or trust abuse.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The manifest description "Operations specialist" is so generic that routing logic may invoke this skill for a wide range of unrelated operational or analytical requests. Overly broad matching can cause unintended skill activation, which increases the chance of inappropriate access to task context, incorrect automation, or delegation to the wrong agent.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The manifest description "Operations specialist" is very broad and non-specific, which increases the chance that routing or invocation systems will select this skill for generic operational, reporting, or administrative requests beyond its intended scope. Overly broad matching can cause inappropriate tool/skill activation, expanding the attack surface for prompt-injection, data exposure, or unintended authority in multi-agent environments.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The description "Operations specialist" is overly generic and overlaps with many common operational or finance-related requests, which can cause the skill to be invoked unintentionally. Over-broad routing increases the chance that a user task is handled by the wrong agent, leading to inappropriate access to operational context, inaccurate outputs, or bypass of more specialized controls.

Content

No source excerpt is available for this finding.