The Agent Skills Directory

[PROMPT_INJECTION]: The skill utilizes highly prescriptive instructions (e.g., "The Iron Law," "Violating the letter... is violating the spirit") to reinforce a specific operational workflow. These instructions serve as behavioral guardrails to ensure verification rigor rather than attempts to bypass agent safety filters or core instructions.
[PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by instructing the agent to ingest and analyze external data, such as reports from other agents and version control (VCS) diffs.
Ingestion points: The skill mentions analyzing agent success reports and VCS diffs in the "Agent delegation" and "Common Failures" sections of SKILL.md.
Boundary markers: The instructions do not define specific delimiters or "ignore" markers for handling these external data sources.
Capability inventory: The skill requires the execution of shell commands through the environment (e.g., uv run pytest, uv run ruff check).
Sanitization: No specific sanitization or escaping mechanisms are provided for the analysis of diffs or reports, which is typical for development-focused automation tasks.
[COMMAND_EXECUTION]: The skill explicitly directs the agent to run verification tools including pytest, ruff, and ty. These commands are standard for Python-based development environments and are used within the context of validating code changes.

pb-verification-before-completion