The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its processing of untrusted data from plan files. It reads markdown files from ai-workspace/plans/ and potentially other skill definitions from .claude/skills/assemble-panel/SKILL.md, passing the extracted content to sub-agents via the Agent tool. The skill lacks boundary markers or explicit instructions to ignore commands embedded within these files, and no sanitization is performed on the ingested content.
[COMMAND_EXECUTION]: The skill executes shell commands to discover and filter plan files in the local workspace. Specifically, Step 1 uses a pipeline of ls, grep, and head to identify the most recent plan file. These are benign, standard utilities used for intended workflow automation within the local environment.

plan-review