The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a local bash script (orchestrate.sh) using a user-provided research question as an argument. While the instructions attempt to wrap the input in double quotes, if the user input contains escaped shell metacharacters (e.g., backticks, dollar signs, or nested quotes) and the underlying shell tool does not provide additional sanitization, it could lead to arbitrary command execution on the host system.
[INDIRECT_PROMPT_INJECTION]: The skill processes and displays external research findings gathered from various providers. This content is untrusted and could contain malicious instructions designed to manipulate the agent's behavior during the synthesis and presentation phase. The skill's instructions acknowledge this risk by referencing a security framing file, but the protection depends on the correct implementation of that external framing.
Ingestion points: User research question (Step 3); synthesis results stored in ~/.claude-octopus/results/ (Step 4).
Boundary markers: Mentions skill-security-framing.md in the security section as a mitigation strategy for external content.
Capability inventory: Full bash shell access via the native shell tool (used to execute orchestrate.sh, find, and cat).
Sanitization: Not explicitly defined within the skill files; relies on external logic or agent-level tool protections.

octopus-research