The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file uses the !command syntax to execute obscura --version at the time the skill is loaded by the agent. While the command itself is benign and used for version checking, this feature allows for immediate code execution in the shell environment.
[EXTERNAL_DOWNLOADS]: The skill instructs the user to download and install the obscura binary from a third-party, non-whitelisted GitHub repository (h4ckf0r0day/obscura). This introduces a dependency on unverified external code.
[COMMAND_EXECUTION]: The scripts/with-obscura.sh script is a utility designed to wrap and execute arbitrary shell commands provided as arguments. This is intended to facilitate running tests with a temporary browser server but represents a capability for executing external processes.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to scrape external web content and convert it to Markdown for agent consumption. This creates a surface where an attacker-controlled website could include malicious instructions designed to manipulate the agent's behavior.
Ingestion points: Untrusted data enters the context via obscura fetch and obscura scrape commands which output page content (SKILL.md, references/cli.md).
Boundary markers: The skill notes an SSRF guard (--allow-private-network is disabled by default) and the Markdown conversion process filters out scripts and styles, though text-based instructions remain (references/cli.md).
Capability inventory: The skill has access to shell execution via with-obscura.sh and network access via the obscura tool itself.
Sanitization: Content is converted to Markdown, but no explicit prompt-level sanitization or 'ignore instructions' delimiters are enforced in the instructions provided to the agent.

obscura