The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill features an attack surface for indirect prompt injection because it processes untrusted data from external sources that may be controlled by third parties.
Ingestion points: Pull Request diffs are fetched and read via gh pr diff in references/code-review-agent.md, and issue content is retrieved using gh issue commands in references/project-manager-agent.md.
Boundary markers: Absent. The sub-agent instructions do not provide delimiters or specific directives to ignore instructions that might be embedded within the external code or issue descriptions.
Capability inventory: The agents possess the ability to execute shell commands (git, gh), post public reviews and comments to GitHub, and modify local repository files such as README.md or CHANGELOG.md.
Sanitization: Absent. There is no explicit logic to validate or sanitize the external data before it is interpolated into the agent's context for decision-making.
[COMMAND_EXECUTION]: The skill heavily relies on shell command execution to perform its primary SDLC functions.
Evidence: The git-workflow-agent.md, code-review-agent.md, and project-manager-agent.md reference files contain numerous shell commands including git checkout, git commit, git push, and various gh (GitHub CLI) subcommands for managing pull requests and issues. The Orchestrator also uses ls for tech-stack detection.

vibe-coder-sdlc