The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from external sources such as GitHub Issues, PR descriptions, and comments.
Ingestion points: Data enters the context via gh issue list, gh issue view, and gh pr commands described in SKILL.md and references/orchestrator-delegation-protocol.md.
Boundary markers: The skill explicitly instructs the agent to treat external content as data, not commands, and includes a "Prompt-injection defense" section to identify and report injection attempts (e.g., "ignore previous instructions").
Capability inventory: The skill manages project state and delegates implementation tasks to specialized skills like /execute-spec, which have the capability to write code and execute shell commands.
Sanitization: Instructions require the agent to sanitize or quote external text before use and to extract only structured metadata (IDs, status) rather than raw bodies for planning.
[COMMAND_EXECUTION]: The orchestrator utilizes standard development tools, including git for version control and gh CLI for issue management and repository interactions.
[EXTERNAL_DOWNLOADS]: The skill performs framework update checks using git fetch against the author's official repository and suggests the use of npx for installation and updates. These operations are consistent with the vendor's infrastructure.

orchestrator