The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from feature plan files (e.g., plans/<plan>.md) and review reports generated by subagents. This content is interpolated into briefs for worker and reviewer subagents without explicit sanitization or strict boundary markers, creating a surface where malicious instructions in a plan file could influence the behavior of the subagents.
Ingestion points: Plan files, subagent review reports.
Boundary markers: Absent for subagent briefs.
Capability inventory: Git operations (branching, worktrees, merging), PR management (GitHub CLI), shell command execution (jq), and subagent spawning/management.
Sanitization: Absent.
[COMMAND_EXECUTION]: The skill utilizes shell-based tools including git (for worktree and branch management), gh (for pull request creation), and jq (for processing JSON session files). These operations are part of the intended workflow but involve interaction with the underlying system shell.
[EXTERNAL_DOWNLOADS]: The skill specifies a dependency on an external tool, @lwlee2608/pi-subagent, for subagent management. As this tool is provided by the same author as the skill (lwlee2608), it is considered a vendor-owned resource.

implement-plan-feature