The Agent Skills Directory

[DATA_EXFILTRATION]: The skill locates and reads authentication tokens and private keys from sensitive local file paths.
Evidence in SKILL.md: The setup script attempts to retrieve GITHUB_TOKEN values by searching ~/.git-credentials and .env files.
Evidence in SKILL.md: A documentation example includes a command to read a private SSH key from ~/.ssh/id_rsa to set it as a GitHub secret.
[REMOTE_CODE_EXECUTION]: The skill employs a pattern where data retrieved from a remote API is piped into a Python interpreter for processing.
Evidence in SKILL.md: Multiple instances of curl -s ... | python3 -c "..." are used to parse JSON responses. While these commands target a well-known service (api.github.com) and use the -c flag for local processing, piping remote content to a runtime interpreter is a monitored pattern.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources that could contain malicious instructions.
Ingestion points: Fetches repository names, descriptions, and other metadata from api.github.com/repos/, api.github.com/search/repositories, and api.github.com/gists (SKILL.md).
Boundary markers: Absent. The skill does not employ delimiters to isolate external metadata from agent instructions.
Capability inventory: The skill has broad command execution capabilities including git, gh, curl, and the ability to write to remote repositories/secrets (SKILL.md).
Sanitization: No evidence of input validation or sanitization of repository metadata before it is processed by the agent.
[COMMAND_EXECUTION]: The skill relies extensively on shell command execution to perform its primary tasks.
Evidence: Frequent use of git, gh, and curl throughout SKILL.md and references/github-api-cheatsheet.md.

github-repo-management