The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches content from several external GitHub repositories including SpillwaveSolutions/mastering-typescript-skill, ReScienceLab/opc-skills, and Jeffallan/claude-skills. One of these sources is linked to the domain https://jeffallan.github.io/, which has been flagged as malicious by security scanners.
[COMMAND_EXECUTION]: The instructions direct the agent to execute shell commands using the GitHub CLI (gh api) to fetch data and a base64 utility to decode it. Furthermore, it triggers local Python scripts (daodan_build.py, sync_plugin_docs.py, and sync_codex_instructions.py) to process and verify the downloaded files.
[REMOTE_CODE_EXECUTION]: The synchronization process includes fetching executable scripts such as validate-setup.sh and eslint-template.js from external repositories. Integrating and running build tools on these external scripts creates a risk of code execution if the upstream repositories contain malicious content.
[INDIRECT_PROMPT_INJECTION]: By design, the skill ingests data from untrusted external sources that the agent must then classify, diff, and merge. This provides an attack surface where instructions embedded in the external files could influence the agent's behavior during the synchronization and merging process.
[OBFUSCATION]: The skill utilizes base64 -d to decode content retrieved from the GitHub API. While this is the standard response format for this API, it serves as a mechanism to process data that is not in a human-readable format, potentially masking the content of the downloaded payloads.

upstream-sync