The Agent Skills Directory

[COMMAND_EXECUTION]: The skill contains multiple examples of shell scripts that use variable interpolation (e.g., $i, $topic, $CONCEPT) directly within command strings. For instance, the 'Instagram Carousel Images' and 'Batch Content Creation' sections use for loops to iterate through values and insert them into belt app run commands. This pattern poses a risk of command injection if these variables were to be populated with untrusted data containing shell metacharacters (e.g., ;, |, `).
[COMMAND_EXECUTION]: The YAML frontmatter specifies allowed-tools: Bash(belt *), which restricts the agent to executing shell commands that begin with the belt prefix. However, many examples in the skill utilize standard bash constructs like variable assignments (TOPICS=(...), CONCEPT=...) and loops (for ... do ... done) that do not begin with belt. This inconsistency may cause execution failures or encourage the agent to attempt to bypass the defined security constraints.
[EXTERNAL_DOWNLOADS]: The skill recommends installing external components including the belt-sh/cli and various other skills hosted in the inference-sh GitHub organization (e.g., ai-video-generation, twitter-automation). It also points to external domains (inference.sh) for documentation and binaries.
[DATA_EXFILTRATION]: The skill documentation includes the belt login command and tools for posting to external platforms (e.g., twitter/post-tweet). These are legitimate functions for the stated purpose of social media management but involve handling user credentials and sending data to third-party services.

ai-social-media-content