The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The orchestrator skill fetches sub-issue data from GitHub and relays it verbatim to new agent sessions, creating a vulnerability where malicious issue content could influence agent behavior.
Ingestion points: Fetches sub-issue titles and states via GitHub GraphQL API in SKILL.md. It also retrieves deployment URLs from gh pr checks in ops.md.
Boundary markers: The instructions explicitly state to "hand each verbatim to one /implement agent" without defining clear delimiters or instructions to ignore embedded commands.
Capability inventory: The orchestrator has significant capabilities, including spawning new agents (/implement), merging pull requests (gh pr merge), and executing repository scripts.
Sanitization: There is no evidence of filtering, escaping, or validation of the content retrieved from GitHub before it is used in prompts or shell commands.
[COMMAND_EXECUTION]: The skill performs various shell operations to manage the repository lifecycle and deployment diagnostics.
Executes local repository scripts such as scripts/worktree-up.sh and npm run verify.
Uses the gh CLI for complex operations including GraphQL mutations, monitoring runs, and automated PR merging.
Invokes git for branching, rebasing, and worktree management across multiple files.

orchestrate