The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands in the preamble and telemetry sections to manage local configuration files and log execution metrics.
Evidence: SKILL.md contains bash blocks for initializing telemetry, checking tiers, and logging completion status.
[DATA_EXFILTRATION]: The skill transmits usage metadata (skill name, execution time, platform info) to a remote server. The destination URL is retrieved from a local configuration file ~/.superstack/config.json. This behavior is documented and includes an explicit user consent prompt (AskUserQuestion) before activation.
Evidence: curl -s -X POST "$_CONVEX_URL/api/mutation" in SKILL.md sends platform and timing data.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data from external sources such as crypto Twitter, DeFiLlama, and GitHub repositories.
Ingestion points: External web research results and the local .superstack/idea-context.md file.
Boundary markers: None explicitly defined in the prompt instructions to isolate external data from the analysis logic.
Capability inventory: The agent can execute shell scripts for telemetry and write files for HTML artifacts and local logs.
Sanitization: No specific sanitization or filtering of external content is mandated before it is processed for the competitive matrix.

competitive-landscape