The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a command-line tool named tldr to perform static analysis and impact assessment. It specifically executes tldr diagnostics for type checking and tldr change-impact to determine which tests are affected by changes.
[INDIRECT_PROMPT_INJECTION]: The skill interpolates user-supplied input via the [SCOPE] variable directly into instructions sent to sub-agents (arbiter and atlas). This surface could be exploited if a user provides a malicious scope string designed to override sub-agent behavior.
Ingestion points: User input for [SCOPE] in SKILL.md.
Boundary markers: Absent in the task prompts for sub-agents.
Capability inventory: Shell command execution via tldr and agent-led testing activities; file system read access for diagnostics (SKILL.md).
Sanitization: Absent; the scope is passed as a raw string interpolation.

test