The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on a suite of bash scripts to automate development workflows. It executes Git commands, GitHub CLI (gh) operations, and Docker commands to manage security fix lifecycles. This includes creating private forks, merging branches, and executing test suites via docker exec.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface in scripts/fetch-advisory.sh and scripts/list-advisories.sh. These scripts retrieve advisory metadata, including summaries and descriptions, directly from GitHub's Security Advisory API. Since this content is provided by external reporters, it could contain instructions designed to mislead the AI agent during the verification or patching phases.
Ingestion points: External data is ingested from the GitHub Security Advisories API via gh api and stored in /tmp/bc-advisories.
Boundary markers: The scripts do not implement specific boundary markers or 'ignore' instructions when presenting advisory content to the agent.
Capability inventory: The agent possesses significant capabilities, including the ability to create forks, push code to remote repositories, and execute commands within a Docker container.
Sanitization: While the Python processing script truncates descriptions to 6000 characters, there is no validation or filtering to prevent embedded instructions from influencing the agent's behavior.

basercms-security-advisory