The Agent Skills Directory

[PROMPT_INJECTION]: The skill utilizes structured subagent roles (Skeptic, Pragmatist, Critic) to provide diverse perspectives. This role-playing is a functional feature for decision-making and does not attempt to override AI safety guardrails.
[DATA_EXFILTRATION]: The skill contains explicit negative constraints preventing the storage of data in shadow directories like ~/.gemini/notes. It directs persistence to official project management tools such as GitHub or Linear, ensuring data remains within authorized boundaries.
[PROMPT_INJECTION]: The skill ingests external data from codebases and metrics, which represents a potential surface for indirect prompt injection. It manages this risk by providing subagents with only compact, relevant context and isolating them from the main conversation history. Evidence chain: (1) Ingestion points: codebase files and snippets collected in SKILL.md Workflow Step 2; (2) Boundary markers: 'Context' and 'Question' headers used in subagent prompts in Step 4; (3) Capability inventory: external writes via knowledge-ops and GitHub/Linear integrations; (4) Sanitization: no explicit filtering of external input is defined.

council