The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill contains shell logic to extract GitHub tokens from local files including ~/.hermes/.env and ~/.git-credentials. While this is functional for an authentication-dependent skill, it highlights a credential access pattern.
[COMMAND_EXECUTION]: The skill uses curl to interact with the GitHub API and pipes the output into python3 -c for processing. This pattern of executing dynamic Python code in a shell environment is common but requires caution if inputs are not properly sanitized.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from GitHub issue titles and bodies (via gh issue view or curl endpoints) and displays them in the agent's context. A malicious issue could contain instructions intended to influence the agent's behavior.
Ingestion points: GitHub issue bodies and titles are fetched from the API and printed to stdout (SKILL.md).
Boundary markers: None identified; external content is printed directly to the console or processed by Python scripts.
Capability inventory: The skill has capabilities to create, edit, close, and comment on issues, as well as execute shell commands and Python scripts.
Sanitization: Content from issues is processed by standard JSON loaders in Python but is not explicitly sanitized for prompt injection before being presented to the agent.

github-issues