The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands for environment detection (ruby -v, rails -v) and file operations (mkdir, test -f). These are standard operations for a development toolkit.
[COMMAND_EXECUTION]: The skill supports executing user-defined commands through phase-inject and verification-gate types. The instructions include a 'Command safety' section in the engine reference that strictly tokenizes these commands and rejects shell metacharacters (|, ;, &&, ||, >, <, backticks, $()) to prevent command injection.
[CREDENTIALS_UNSAFE]: The skill handles private gem sources (e.g., Rails LTS) by referencing environment variables (credentials_env). The engine reference explicitly mandates that credential values are never stored or logged, only their presence is checked. This follows best practices for secret management.
[DYNAMIC_EXECUTION]: The skill supports regex-based code transformations. To mitigate ReDoS attacks, the engine reference requires a catastrophic-backtracking check using safe-regex heuristics and validates patterns as Ruby regexes before authoring.
[INDIRECT_PROMPT_INJECTION]: The skill reads a .ruby-upgrade-toolkit/rules.yml file from the project root. While this is an ingestion point for external data, the skill implements a strict schema validation pipeline (described in schema.md) that restricts inputs to known enums and whitelisted settings, minimizing the risk of indirect injection.

Custom Rules