The Agent Skills Directory

[COMMAND_EXECUTION]: The lib/backend_selector.py script implements a 'Tier 2' publishing mechanism that utilizes subprocess.run() to execute a command string defined in the THREADS_SKILLS_CUSTOM_POSTER environment variable. This allows the execution of arbitrary shell commands within the agent's environment.
[DYNAMIC_EXECUTION]: The skill instructions explicitly encourage the AI agent to 'build a custom poster' and set the execution path in an environment variable at runtime. This pattern of dynamic command construction and execution significantly increases the attack surface, as the agent's logic for building the command could be subverted by malicious input.
[INDIRECT_PROMPT_INJECTION]: The threads-audience-insights and threads-hook-extractor skills process untrusted external data retrieved via the Apify API (e.g., niche posts, competitor bios) or pasted from the user's clipboard. This data is used to extract strategies and draft content.
Ingestion points: ApifyClient fetches remote content from the Threads platform; multiple skills accept user-pasted text as primary input for analysis.
Boundary markers: The skill bundle implements a 'human-in-the-loop' approval card via lib/approval.py, requiring explicit user consent before any action is taken. However, the initial processing of untrusted data occurs before this boundary.
Capability inventory: The agent has network access via the requests library, file system access to manage the voice-profile.md file, and shell execution capabilities through the DIY poster feature.
Sanitization: The threads-humanizer skill provides style-based scrubbing for AI tells (e.g., removing em-dashes and specific vocabulary), but does not implement security-focused sanitization or instruction filtering for incoming social media data.

threads-marketing