The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external data from GitHub issue bodies, PR comments, and CI failure logs to guide sub-agents in generating code fixes. This represents an indirect prompt injection surface.\n
Ingestion points: GitHub CLI commands gh issue list, gh run view --log-failed, and gh issue view --json comments (in SKILL.md and resources/comment-policy.md).\n
Boundary markers: None identified; untrusted data appears to be passed directly to sub-agents.\n
Capability inventory: Sub-agents can write to worktrees and execute shell commands (npm ci, npm test).\n
Sanitization: No sanitization or validation of external content is specified before interpolation into agent prompts.- [REMOTE_CODE_EXECUTION]: The skill initiates execution of code generated by sub-agents using npm ci, npm test, and npm run build within a git worktree. If the code generation process is compromised via indirect prompt injection, these commands could execute malicious logic embedded in the repository (e.g., via package.json lifecycle hooks in package.json).- [COMMAND_EXECUTION]: The skill uses shell-based tools (gh, git, npm) to manage the repository and interact with the GitHub API. These commands are executed based on the current state of the repository and findings.- [DATA_EXFILTRATION]: The skill uses the GitHub CLI (gh) to transmit information to and from the bitbonsai/mcpvault repository. While this is the core functionality, the network access could potentially be misused for exfiltration if the agent's instructions were overridden.

triage