The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructional shell script templates for automating developer tasks using the claude CLI and GitHub (gh) CLI. These are standard automation patterns for the intended developer-focused use case.
[PROMPT_INJECTION]: The skill describes architectures that ingest external data (specification files, CI logs via gh run view, and project documentation) into agent prompts. This represents a standard indirect prompt injection surface for agentic workflows.
Ingestion points: Markdown specification files (specs/*.md) and CI failure logs fetched via GitHub CLI.
Boundary markers: The provided templates do not specify explicit delimiters or instruction-guarding prompts for external data, though the 'Ralphinho' pattern suggests separating context windows to eliminate author bias.
Capability inventory: The loop implementations utilize the claude CLI, which has file system access and shell execution capabilities via the host agent tools (Read, Write, Edit, Bash).
Sanitization: No explicit sanitization or validation of the ingested external content is included in the reference patterns, relying on the agent's internal safety filters.

autonomous-loops