The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content retrieved from X (Twitter), creating an attack surface for malicious instructions to influence the agent.
Ingestion points: Tweet text, handles, and metadata fetched via the X.AI API or WebSearch tools (e.g., in the agent-buzz and topic branches) are parsed and used to generate curated digests and insights in SKILL.md.
Boundary markers: The skill does not employ explicit delimiters or system-level instructions to ignore embedded commands within the fetched tweet content when the agent performs summarization or clustering.
Capability inventory: The agent has the ability to write to local memory files (memory/, MEMORY.md), perform network operations via curl, and send notifications to the user.
Sanitization: There is no evidence of sanitization or filtering logic designed to detect or neutralize adversarial prompt injection attempts embedded within social media posts before they are processed by the agent's logic.
[COMMAND_EXECUTION]: The skill dynamically constructs and executes shell commands to interact with external APIs and local utilities.
Evidence: The skill uses curl (via a ./secretcurl wrapper) and jq to build and send API requests to X.AI. Variables such as ${ARG}, ${LIST_ID}, and cleaned handles like ${ACCOUNT} are interpolated into these shell commands.
Mitigation: The skill includes validation steps, such as regex checks for numeric list IDs and normalization for account handles, which reduces but does not entirely eliminate the risks associated with dynamic command generation.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to external domains to fetch tweet data.
Evidence: It communicates with the X.AI API at api.x.ai and uses WebSearch/WebFetch tools to access x.com and twitter.com. These are well-known services relevant to the skill's primary function.

fetch-tweets