The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's 'Panel mode' executes shell commands using python3 to run scripts located at shared/scripts/panel_round.py and shared/scripts/validate_artifacts.py. These scripts take arguments such as --goal which are derived from user-provided plans or PRDs, potentially allowing for argument injection if the input is not properly sanitized.
[DYNAMIC_EXECUTION]: The assets/panel-routing.toml configuration file defines a complex infrastructure for 'runners' that execute external scripts (.agents/skills/claude-runner/scripts/run_claude.py, .agents/skills/gemini-runner/scripts/run_gemini.py, etc.). This runtime orchestration of multiple agents via shell scripts represents a significant dynamic execution surface.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external specifications (PRDs, plans, CONCEPTS.md, ADRs). These inputs directly influence the generation of 'Slice Contracts' and task instructions for downstream agents. A malicious PRD could include hidden instructions to weaken security gates or bypass testing requirements in the resulting task files.
Ingestion points: Reads .ai-workflow/work/<feature-slug>/prd.md, CONCEPTS.md, and arbitrary plan files or URLs provided by the user (SKILL.md).
Boundary markers: The skill does not define specific delimiters or 'ignore' instructions for the content it reads from the codebase or PRDs.
Capability inventory: The skill writes multiple markdown files to the filesystem (.ai-workflow/work/<feature-slug>/tasks/) and executes external Python scripts via shell commands.
Sanitization: No explicit sanitization or filtering of the ingested specification content is mentioned before it is used to generate tasks.

to-tasks