The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/bluesky_analyzer.py constructs a Python script as a string and executes it via subprocess.run using a specific virtual environment interpreter. This technique is used to delegate keyword extraction tasks to the yake library within that environment by passing the script via the -c flag.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external Bluesky accounts including user bios and post content. This data is processed and presented to the AI agent for categorization, creating an attack surface where malicious instructions embedded in Bluesky content could influence the agent's behavior.
Ingestion points: scripts/bluesky_analyzer.py fetches data via the Bluesky API using the app.bsky.graph.getFollows and app.bsky.feed.getAuthorFeed endpoints.
Boundary markers: The output uses simple text labeling like @handle and Keywords: but lacks explicit security delimiters or instructions to ignore embedded commands.
Capability inventory: The skill uses subprocess.run in scripts/bluesky_analyzer.py to execute generated Python code.
Sanitization: No sanitization or filtering is performed on the bio or post content before it is processed by the keyword extractor or presented to the agent.
[COMMAND_EXECUTION]: The script uses the subprocess module to execute internal Python commands. While used for its primary functional purpose, it represents a shell-level execution capability within the skill.

categorizing-bsky-accounts