The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple sources to synthesize or update agent instructions. This creates a surface where malicious content embedded in source files, transcripts, or repositories could influence the agent during the synthesis phase.
Ingestion points: The skill ingests data from external URLs in ADAPT mode (SKILL.md), video transcripts in CREATE/from-video mode (SKILL.md), pasted notes in CREATE/from-dump mode (SKILL.md), and recent conversation history in UPDATE/from-chat mode (SKILL.md).
Boundary markers: No explicit boundary markers or instructions to disregard embedded commands were identified in the instruction set for processing external content.
Capability inventory: The skill possesses the ability to write to the file system using the Edit tool, perform Git operations (clone, commit, push), and interact with the GitHub API via the gh CLI.
Sanitization: While a license check is performed during adaptation, the skill does not include specific sanitization or filtering logic to detect or neutralize prompt injection attempts within the source data.
[EXTERNAL_DOWNLOADS]: The skill fetches external content to facilitate the adaptation of remote skills.
In SKILL.md (ADAPT mode), the skill retrieves skill definitions using git clone and the GitHub API (gh api) from user-provided URLs targeting GitHub and other skill registries.
The references/adapt-license-check.md utility uses curl to fetch repository license metadata from the official GitHub API.

skillify