The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes npx to dynamically download and run packages such as ruv-swarm and claude-flow from the NPM registry. These are vendor-owned tools associated with the skill author.
[REMOTE_CODE_EXECUTION]: Remote code execution is performed via npx for executing swarm coordination logic and through the reference to ruvnet/swarm-action@v1 for GitHub Actions automation.
[COMMAND_EXECUTION]: The skill relies on extensive Bash command execution using the GitHub CLI (gh) and npx to perform repository management, metadata manipulation, and project board automation.
[DATA_EXFILTRATION]: The skill documentation includes examples of transmitting project data to external endpoints, such as a configurable --webhook-endpoint and distribution of reports via slack or email.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting and processing content from GitHub issues (titles, bodies, and comments) which can be influenced by external actors.
Ingestion points: Untrusted issue data is retrieved through gh issue view and gh issue list commands, which populate variables like $ISSUE_DATA and $ISSUE_BODY in the provided workflows.
Boundary markers: The scripts do not implement specific boundary markers or instructions to the AI to ignore embedded commands within the ingested issue content.
Capability inventory: The skill possesses significant write capabilities, including editing issues, posting comments, and spawning AI agents with specialized roles via mcp__claude-flow__agent_spawn.
Sanitization: No explicit sanitization, validation, or filtering of the retrieved issue content is performed before it is passed to the swarm agents for processing.

github-project-management