Security audit
Security checks for vulnerabilities and agentic risk
This skill does what it says: it uploads HTML or Markdown to a third-party public sharing service, with the main risk being accidental public disclosure if used on sensitive content.
Install only if you want agents to create public DropFiles links. Do not use it for secrets, credentials, private reports, regulated data, or internal-only documents unless you intentionally approve that upload and use password protection when appropriate.
Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.
Use the bundled wrapper script. Do NOT construct curl commands or JSON payloads manually.
python3 scripts/upload.py [OPTIONS]
Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.
"""DropFiles upload wrapper.
Uploads HTML/Markdown content to DropFiles and returns a shareable link.
Encapsulates all curl / JSON / temp-file / WAF-workaround logic so callers
(agents or humans) only need one Shell invocation.
Usage:
The skill recommends invocation for broad situations like 'share' or 'publish' generated content, which can cause an agent to upload material externally in many ordinary workflows without a sufficiently explicit user intent to make content public. In this skill’s context, unintended activation is more dangerous because the action sends potentially sensitive generated files to a third-party public hosting service where links are public by default unless password protection is manually enabled.
The skill invokes local file reads, shell execution, and outbound network access via python3 scripts/upload.py, but it does not declare any tool scope or permission boundaries. This increases the chance that an agent will run the skill with broader capabilities than intended, making unintended data exposure or command execution harder to constrain or audit.
The description says to invoke the skill whenever the user wants to 'share, publish, or generate a public link' for produced content, which is broad enough to match many ordinary requests. Over-broad routing can cause an agent to upload sensitive generated artifacts to a public external service when the user may only have intended local sharing or an internal handoff.
The 'When to Invoke' triggers are very general and lack boundary conditions, so routine requests to share or preview content may trigger an external upload automatically. In the context of a skill that publishes content to a public link by default, this ambiguity materially raises the risk of accidental disclosure.
The script transmits arbitrary HTML/Markdown content to a third-party remote service and is designed for agent automation, but it provides no explicit user-facing consent or warning at execution time about external data disclosure. In an agent context, this increases the chance that sensitive generated reports, secrets, or internal content are uploaded publicly without the user fully realizing the data is leaving the local environment.
subprocess module calls execute external commands. Without careful input validation, this enables command injection.
"-d", f"@{tmp_path}",
"-w", "\n%{http_code}",
]
result = subprocess.run(cmd, capture_output=True, text=True)
if result.returncode != 0:
fail(