The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes common developer tools including the GitHub CLI (gh), git, and npm to perform its tasks.
The npm view command is executed with a temporary configuration (--userconfig "$(mktemp)"), which is a security best practice to avoid interacting with sensitive local configuration files.
The pbcopy command is used to copy drafted responses to the system clipboard, requiring the user to manually paste and post the content.
[DATA_EXFILTRATION]: The skill retrieves information from GitHub's security advisory and issue APIs. While this involves bringing external data into the agent's context, the operations are limited to the specified openclaw/openclaw repository. No sensitive local data is transmitted to unauthorized domains.
[PROMPT_INJECTION]: The skill processes data from external security advisories and GitHub issues, which presents a surface for indirect prompt injection.
Ingestion points: External data is fetched using gh api and gh search commands from the openclaw/openclaw repository.
Boundary markers: No explicit delimiters are used to separate the system instructions from the content of the fetched advisories.
Capability inventory: The skill uses shell commands for repository management and clipboard interaction.
Sanitization: No specific sanitization or filtering is implemented for the content of the security advisories before they are summarized or copied.
Mitigation: The potential risk is effectively mitigated by the operational requirement for a human maintainer to review and manually post any generated content.

security-triage