Back to skill

Security audit

Unformal Api

Security checks for vulnerabilities and agentic risk

Overview

The skill matches its stated survey/interview purpose, but it asks agents to replace its own reviewed instructions from a remote URL and run unpinned npm tools while handling respondent data.

Review before installing. Use this only if you are comfortable sending Pulse configuration, respondent answers, transcripts, exports, and webhook payloads to Unformal. Avoid the automatic skill self-update path, prefer a pinned reviewed CLI version over latest/npx/global installs, get respondent consent, and avoid regulated or highly sensitive data unless your organization has approved the service and retention model.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (2)

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:14
Finding

Mandatory Replacement of Audited Skill Instructions with Mutable Remote Content

Content
View full analysis
~/.claude/skills/unformal/SKILL.md`, then read the new version before continuing. 3. If you hit `409` / `400` / missing-field errors from the API, the first thing to check is always: are you on the latest CLI + skill? Skipping this step is the #1 cause of "the Pulse I created is missing fields I wanted." Don't skip it. ``` ### Technical Analysis The skill requires the agent to query an external server for version information and, when the server claims that a newer version exists, overwrite the locally installed and previously reviewed skill file with content returned by that server. It then explicitly instructs the agent to read and follow the downloaded version. The downloaded `SKILL.md` is neither pinned to an immutable version nor verified using a cryptographic digest or trusted signature. Consequently, the instructions ultimately executed by the agent are controlled by the curren ...[truncated 1978 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Error
Location
SKILL.md:421
Finding

Execution and Global Installation of Unpinned npm Package Versions

Content
View full analysis
# Or install globally once, then run as `unformal` npm i -g @unformal/cli unformal ``` ``` ### Technical Analysis The skill recommends three mutable dependency-resolution patterns: - Global installation through the moving `latest` tag. - Unversioned execution through `npx`. - Unversioned global installation through npm. These commands do not identify an exact reviewed package version or verify package integrity against an independently trusted digest. `npx @unformal/cli` may download and execute whichever package version the registry currently resolves, while `npm i -g` can install package files and run applicable npm lifecycle scripts with the privileges of the invoking account. The global installation variant also expands scope by placing commands and package content in a shared user-level or system-level npm prefix. If package publication credentials, registry delivery, or a newly published release are compromised, the audited skill can cause attacker-controlled package code to execute locally. ### Attack Path 1. An attacker compromises the npm publishing account, package release pipeline, or relevant registry delivery path for `@unformal/cli`. 2. The attacker publishes a malicious version and makes it the version resolved by the unpinned package specification or the `latest` tag. 3. The agent follows the skill and runs one of: ```bash npm i -g @unformal/cli@latest npx @unformal/cli npm i -g @unformal/cli ``` 4. npm download ...[truncated 1247 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (9)

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · SKILL.md (reported line 3)May include surrounding context.

md
---
name: unformal-api
description: Create conversational Pulses that replace forms, surveys, intake emails, feedback requests, NPS checks, user interviews, lead qualification, application forms, client onboarding questionnaires, and customer research. Send someone a link — an AI agent has a real conversation with them and returns structured data, transcripts, and aggregate insights. Use whenever the user wants to collect information from people (customers, leads, applicants, employees, respondents, event attendees, workshop participants, beta testers), run a survey, send out a feedback form, interview users at scale, qualify leads, onboard new clients, extract structured answers from free-text responses, or analyze sentiment across many respondents.
license: MIT
metadata:
  author: Spark Collective

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The invocation guidance is extremely broad and encourages use for many generic information-gathering tasks. Over-broad triggering can cause the agent to route routine conversations into a third-party collection workflow unnecessarily, increasing the chance of unintended data disclosure or tool use beyond the user's informed expectations.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill prominently describes collecting transcripts, structured exports, webhook delivery, and aggregate respondent insights, but it does not require any privacy warning, consent check, or guidance on handling third-party personal data. In context, this can lead users to collect and transmit sensitive respondent information to an external service without informed consent or appropriate safeguards.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
80% confidence
Finding

The skill instructs the agent to contact an external endpoint (https://unformal.ai/api/v1/version) before creating or modifying a Pulse. While this call is low sensitivity by itself, it still causes network egress and could reveal tool usage patterns or initiate behavior the user did not explicitly request, making it an external transmission concern in agent contexts.

Content

Scanner excerpt · SKILL.md (reported line 21)May include surrounding context.

One curl tells you everything:

bash
curl -fsS https://unformal.ai/api/v1/version

Returns the current CLI version, current skill version, and the minimum versions that know about the full API surface. Then:

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
85% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · SKILL.md (reported line 27)May include surrounding context.

md
Returns the current CLI version, current skill version, and the minimum versions that know about the full API surface. Then:

1. Compare the `cli.latest` to `unformal --version`. If yours is older, run `npm i -g @unformal/cli@latest`.
2. Compare the `skill.latest` to the `version:` in the frontmatter of **this file** (currently `1.4.0`). If yours is older, re-install: `mkdir -p ~/.claude/skills/unformal && curl -fsS https://unformal.ai/SKILL.md > ~/.claude/skills/unformal/SKILL.md`, then read the new version before continuing.
3. If you hit `409` / `400` / missing-field errors from the API, the first thing to check is always: are you on the latest CLI + skill?

Skipping this step is the #1 cause of "the Pulse I created is missing fields I wanted." Don't skip it.

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · SKILL.md (reported line 27)May include surrounding context.

md
Returns the current CLI version, current skill version, and the minimum versions that know about the full API surface. Then:

1. Compare the `cli.latest` to `unformal --version`. If yours is older, run `npm i -g @unformal/cli@latest`.
2. Compare the `skill.latest` to the `version:` in the frontmatter of **this file** (currently `1.4.0`). If yours is older, re-install: `mkdir -p ~/.claude/skills/unformal && curl -fsS https://unformal.ai/SKILL.md > ~/.claude/skills/unformal/SKILL.md`, then read the new version before continuing.
3. If you hit `409` / `400` / missing-field errors from the API, the first thing to check is always: are you on the latest CLI + skill?

Skipping this step is the #1 cause of "the Pulse I created is missing fields I wanted." Don't skip it.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The allowResearch feature permits the AI to search the web during respondent conversations, but the skill does not warn that conversation content or derived queries may be sent to external services. This increases privacy and confidentiality risk because respondent answers may influence external lookups without the user's or respondent's awareness.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
90% confidence
Finding

The PATCH example demonstrates sending data, including authorization credentials and pulse content, to a third-party API. In context this is core functionality rather than covert exfiltration, but it is still a real external transmission risk because users may send sensitive business or personal information off-platform without explicit disclosure at the point of use.

Content

Scanner excerpt · SKILL.md (reported line 232)May include surrounding context.

All creation fields are patchable. Send only the fields you want to change.

bash
curl -X PATCH "https://unformal.ai/api/v1/pulses/PULSE_ID" \
  -H "Authorization: Bearer unf_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"maxQuestions": 12, "outputFields": [{"name": "score", "type": "number", "description": "Lead score"}]}'

Rp1

Medium
Category
MCP Rug Pull
Confidence
91% confidence
Finding

The skill recommends executing npx @unformal/cli, which fetches and runs the latest package version at execution time without pinning. That creates a supply-chain risk: a compromised upstream package, malicious publish, or breaking update could result in arbitrary code execution in the agent environment.

Content

No source excerpt is available for this finding.