The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted ticket content and interpolates it into subagent prompts, creating a vulnerability surface for embedded secondary instructions.\n
Ingestion points: Untrusted data enters the agent context from local markdown files in .scratch/ or remote trackers (e.g., GitHub, Linear) as defined in docs/agents/issue-tracker.md (see references/ticket-lifecycle.md).\n
Boundary markers: Boundary markers are absent; the references/worker-contract.md template uses a simple text header and {full ticket markdown} placeholder without robust delimiters (like XML tags) or explicit instructions to the subagent to ignore embedded commands.\n
Capability inventory: The orchestrator and its workers execute subprocesses via git, run project tests, and perform file system writes during the integration phase as seen in references/host-adapters.md.\n
Sanitization: Sanitization is absent; the skill does not define any processes for escaping or validating ticket content before it is passed to the subagent's execution environment.\n- [COMMAND_EXECUTION]: The orchestrator performs extensive shell operations for repository management and testing.\n
Evidence: Uses git worktree, git log, git rev-parse, and git status to manage ticket isolation and verify completion states across the orchestrator and worker worktrees.\n
Evidence: Invokes project-specific tests and coordinates the execution of external skills (/code-review, /implement, /tdd) on the host system.

implement-with-subagents