Back to skill

Security audit

Everyday Help Activator

Security checks for vulnerabilities and agentic risk

Overview

This skill is not overtly malicious, but it should be reviewed because it asks the agent to keep ongoing local notes about real people without clear consent, limits, or deletion guidance.

Install only if you are comfortable with the agent creating local records about real-world interactions. Avoid names, identifying details, reactions, locations, and service-worker profiles; keep any notes out of shared or synced folders; delete them regularly; and prefer anonymized, opt-in summaries. Also verify the ClawHub install source because the npx command is not pinned.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (2)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:31
Finding

Persistent Plaintext Profiling of Real-World Interactions

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 31–46
Vulnerability Type: Plaintext storage of potentially identifiable personal and behavioral data
Risk Level: Medium

Complete Code Snippet:

markdown
**Agent action**: In your head or notes app, run the 3-second checklist: Capable? Tiny ask? Genuine opener ready? If yes, proceed. Use the filesystem to log successful activations in `help-log.md` (date, person type, ask, outcome) to track what works in your area.

### Step 2: Deliver the Micro-Request Script
Use this exact 4-part formula out loud (keep it under 15 seconds total):
1. Genuine opener (smile + eye contact): “Hi, I really like how [specific positive thing] you did.”
2. State the tiny need: “I’m trying to [simple action] and could use a quick hand for literally 10 seconds.”
3. Make it easy for them: “Would you mind just [super-specific one-step action]?”
4. Immediate gratitude + exit: “Thank you so much — you just saved me.”

**Agent action**: Practice the full script aloud once before approaching (record yourself if possible). After the interaction, immediately note in `help-log.md`: exact script used, their reaction (warm / neutral / hesitant), and what you offered in return.

### Step 3: Close with Reciprocity (The Glue)
Within 10 seconds of their help, offer something back — even if tiny: “Next time I see you I’ll grab the door for you” or hand them a spare bottle of water / snack you have, or simply “If you ever need anything like this, I’m around.”

**Agent action**: If it’s a repeat person (neighbor, regular cashier), create a new file `relationship-[name-or-type].md` and note the date + small reciprocity offered. Update it each time you see them to keep the ally warm.

Technical Analysis

The skill instructs the agent to persist dates, person types or names, requests, reactions, and relationship histories in ordinary Markdown files. These records may identify or make rec ...[truncated 1652 chars]

Remediation
View remediation

Remediation Suggestions

  • Make interaction logging optional and disabled by default.
  • Store anonymous aggregate statistics instead of person-level records.
  • Explicitly prohibit names, contact information, precise locations, physical descriptions, work schedules, and other identifying details.
  • Do not create one file per person; use non-identifying counters or ephemeral session notes.
  • Define a short retention period and automatically delete expired records.
  • Require informed consent before retaining information about another person.
  • Keep any necessary records outside shared repositories and apply restrictive filesystem permissions.
  • Warn users not to synchronize, publish, or commit the records to source control.
  • Avoid recording audio unless all affected parties have consented and applicable law permits it.

T08 · Insecure Dependencies

Warning
Location
SKILL.md:11
Finding

Unpinned Package Execution Through npx Installation Command

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 11
Vulnerability Type: Mutable third-party package execution without version or integrity pinning
Risk Level: Medium

Complete Code Snippet:

yaml
install: "npx clawhub install everyday-help-activator"

Technical Analysis

The installation metadata invokes npx clawhub without specifying an exact package version or integrity digest. Depending on the local environment and npx behavior, the command may retrieve and execute the currently published clawhub package from a configured package registry.

Because the effective package content can change after this skill has been audited, the reviewed repository does not fully determine installation behavior. Compromise of the package publisher, registry account, dependency chain, package resolution configuration, or selected release could introduce arbitrary executable behavior. The project contains no lockfile, checksum, trusted publisher identifier, or other mechanism that binds installation to a reviewed artifact.

This finding does not establish that the current clawhub package is malicious. The weakness is that the command permits mutable third-party code to be selected and executed without cryptographic verification.

Attack Path

  1. An attacker compromises the package publisher, registry account, distribution channel, or a dependency used by the resolved clawhub release.
  2. The attacker publishes or causes resolution of a modified package version containing malicious CLI or lifecycle code.
  3. A user or installation system executes npx clawhub install everyday-help-activator.
  4. npx resolves and downloads the mutable package content from the configured registry.
  5. The malicious package code executes with the privileges and environment of the user running the installation command.
  6. The code may access files, credentials available to that process, network resources, or the instal ...[truncated 526 chars]
Remediation
View remediation

Remediation Suggestions

  • Pin clawhub to an exact, reviewed version rather than resolving the latest available release.
  • Verify the package using a trusted integrity hash or signed provenance before execution.
  • Document the expected registry and verified publisher identity.
  • Use a lockfile or equivalent immutable dependency manifest where supported.
  • Disable or review package lifecycle scripts during installation.
  • Prefer a preinstalled, centrally managed CLI or a locally audited artifact over automatic remote execution.
  • Run installation in a sandbox with minimal filesystem, credential, and network access.
  • Establish a controlled process for reviewing and updating the pinned version.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (8)

Vague Triggers

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The description says to use the skill 'whenever you need an extra hand in real life' and lists very common situations like directions, borrowing tools, and emergency assistance. This makes the trigger scope extremely broad and likely to collide with ordinary conversation or many unrelated help-seeking contexts.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

Directing the agent to create persistent relationship files about people, including neighbors, cashiers, or maintenance staff, without user warning materially increases privacy risk. In context, the skill is about influencing real-world social interactions, so adding durable records about specific people makes the behavior more intrusive and potentially manipulative.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The manifest instructs installation via npx clawhub without pinning a specific version, which makes execution dependent on whatever package version is current at install time. That creates a supply-chain risk: a compromised or malicious upstream release could alter behavior or introduce unsafe code without the skill author changing this file.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The listed conditions cover many everyday encounters such as seeing someone nearby, asking service staff for exceptions, or starting reciprocal relationships with regular contacts. The file does not provide clear limits or negative examples to distinguish appropriate activation from ordinary social situations, making invocation criteria ambiguous.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill tells the agent to create help-log.md entries about interactions with identifiable categories of people, scripts used, and reactions, without any privacy warning or minimization guidance. This can normalize silent collection of interpersonal data in local files, which may later be exposed, synced, or reused beyond the original purpose.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The instructions create a natural-language data collection workflow covering encounters, reactions, reciprocity, and repeat-person tracking. Even if stored locally, this is a structured retention pattern about third parties that can enable profiling, mission creep, and accidental disclosure, especially because the people being logged are not the user and are unlikely to be aware of the recordkeeping.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill is presented as conversational help-request guidance, but it also directs persistent logging of interactions with nearby people, including person type, reactions, and outcomes. This expands the skill from ephemeral coaching into behavioral data collection about third parties, creating privacy and retention risks that users may not expect.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
94% confidence
Finding

The instruction to create relationship-[name-or-type].md files for repeat people encourages building persistent profiles on identifiable individuals unrelated to the core task of requesting small favors. Maintaining per-person relationship files increases the chance of unnecessary personal data accumulation, misuse, and long-term tracking of acquaintances or service workers.

Content

No source excerpt is available for this finding.