The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill downloads and executes a binary named eyebrow from a third-party GitHub repository (alexverify/eyebrow).
Evidence: The script in step S7 uses gh release download to fetch the binary and executes it using env -i to generate an integrity lock for new skills.
[EXTERNAL_DOWNLOADS]: Fetches framework updates and binary assets from external GitHub repositories.
Evidence: Uses gh api to compare and fetch content from an upstream repository (defaulting to aeonfun/aeon but overridable via the repo= parameter) and gh release download for the eyebrow tool.
[COMMAND_EXECUTION]: Employs shell commands to perform repository mutations and GitHub interactions.
Evidence: Uses git merge-file for 3-way merges, git commit to stage changes, and gh pr create to submit updates for review.
[PROMPT_INJECTION]: The skill processes untrusted data from an external upstream repository, creating a surface for indirect prompt injection.
Ingestion points: File contents and commit messages are fetched from the UPSTREAM repository via GitHub API in SKILL.md (S4, S6).
Boundary markers: The PR body template uses Markdown headers and specific sections to isolate upstream content (S9).
Capability inventory: The skill can write files, manage git branches, create Pull Requests, and execute binaries (S7, S9).
Sanitization: It uses git commit -F to handle commit messages safely, avoiding direct shell interpolation of commit text (S9).

aeon-update