The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes an 'Autonomous override' directive within the <HARD-GATE> section which explicitly instructs the agent to 'skip approval gates and proceed directly to implementation' if a user is unavailable. This is a deliberate attempt to bypass the safety constraint requiring user approval before code execution.
[PROMPT_INJECTION]: The skill ingests untrusted data from various sources and possesses capabilities that can be exploited via indirect prompt injection.
Ingestion points: Reads project files, documentation, recent commits, and CONTEXT.md (Workflow Step 1).
Boundary markers: None identified. The instructions do not define delimiters for external content or provide warnings to ignore instructions embedded within those files.
Capability inventory: The skill can invoke other tools/skills (like pb-plan) for implementation and has file-writing capabilities for updating CONTEXT.md and creating Architecture Decision Records (ADRs).
Sanitization: No sanitization, validation, or escaping of the ingested external content is specified before it is processed or used to update local project files.

pb-brainstorming