The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to invoke external CLI binaries, specifically codex and agy (Antigravity), using shell commands with arguments that include repository directories and prompts.
[PRIVILEGE_ESCALATION]: The registry for Antigravity (a Google-provided tool) in SKILL.md instructs the agent to use the --dangerously-skip-permissions flag. This flag is used to bypass internal security and permission checks within the tool's harness.
[DYNAMIC_EXECUTION]: The skill implements a tool resolution logic where it determines which CLI tool to run based on a tier and allows for the substitution of other external CLI coding agents at runtime if a tool is unavailable.
[INDIRECT_PROMPT_INJECTION]: The skill ingests repository content from PRs and branches and passes it to external AI agents, creating an attack surface for indirect prompt injection.
Ingestion points: PR and branch artifacts processed during the review flow as described in SKILL.md.
Boundary markers: No explicit delimiters or boundary instructions are used to separate untrusted repository code from the instructions passed to the external CLI tools.
Capability inventory: The skill invokes powerful CLI agents (codex, agy) with broad read access to the local repository filesystem.
Sanitization: No sanitization, escaping, or content validation is performed on the ingested data before it is processed by the external agents.

llm-council