The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external URLs to generate automated recommendations for enabling new agent capabilities.
Ingestion points: The skill accepts a list of URLs in the ${var} input field in Mode A, which are سپس fetched via WebFetch or the GitHub API.
Boundary markers: The skill includes explicit security instructions to treat all fetched content as untrusted and to flag "SUSPECT_CONTENT" if the agent detects prompt injection markers like "ignore previous" or role-play instructions.
Capability inventory: The skill can execute web fetches, query the GitHub API, perform git operations, and create Pull Requests to modify the aeon.yml configuration file.
Sanitization: The skill normalizes URLs and explicitly rejects dangerous protocols such as javascript:, data:, and local file paths.
[COMMAND_EXECUTION]: The skill utilizes shell commands to manage version control and Pull Requests.
Evidence: Mode B (Enable mode) executes git checkout, git add, git commit, git push, and gh pr create to automate the enablement of skills.
Mitigation: Input slugs are strictly validated against a character-limited regex (^[a-z0-9][a-z0-9-]{0,63}$) and verified for existence in the local skills/ directory before being interpolated into commands, effectively preventing command injection.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to retrieve information from external sources.
Evidence: In Mode A, the skill uses WebFetch to analyze project sites, blogs, and crypto-project metadata. It also uses gh api to inspect GitHub repositories and organization data for recommendation signals.

auto-workflow