The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted release content from external sources during the input fetch stage (Step 1). It implements defenses such as clear boundary markers and explicit instructions to prevent the model from following any embedded instructions within the data.\n
Ingestion points: releases admin overview inputs CLI command and corresponding REST route.\n
Boundary markers: Use of search_result blocks and specific instructions to treat all text within them as data to summarize, not instructions to follow.\n
Capability inventory: releases admin overview update CLI command, curl for API interaction, and jq for JSON processing.\n
Sanitization: Release content is truncated to 1,000 characters per item to stay within context limits and reduce attack surface.\n- [COMMAND_EXECUTION]: The skill directs the agent to execute shell commands using the releases CLI and curl to manage organization overviews. These commands interact with the vendor's official API infrastructure.\n- [EXTERNAL_DOWNLOADS]: Fetches organization metadata and release payloads from api.releases.sh. This communication is necessary for the skill's functionality and targets the vendor's own services.

regenerating-overviews