The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a inherent attack surface for indirect prompt injection due to its core functionality.
Ingestion points: The workflow fetches and processes external Jira tickets and GitHub issues in Phase 1, storing them in local documentation files (docs/<KEY>.md) for use in planning and execution phases.
Boundary markers: While the artifact-validator subagent checks for structural integrity and specific markdown headers, there is no mechanism to delimit or ignore instructions that may be embedded within the body of the fetched work items.
Capability inventory: The orchestrator and its subagents have broad read/write capabilities, including repository searches (rg), git state inspection, platform writes via the gh CLI or Jira MCP, and the ability to invoke other runtime skills.
Sanitization: The skill does not explicitly sanitize or filter natural language instructions found in external data before they are used to influence downstream planning (Phase 2) or task execution (Phase 7).
[EXTERNAL_DOWNLOADS]: The skill references and directs the agent to fetch information from third-party domains not recognized as trusted official sources.
Evidence: The references/external-sources.md file contains URLs pointing to agentskills.io and skills.sh.
Usage: These external resources are provided as sources for the agent to fetch and summarize conceptual background or setup instructions for dependencies, which could introduce untrusted guidance into the agent's context.

orchestrating-workflow