The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and synthesize information from potentially untrusted external sources, which could be used to manipulate the instructions passed to sub-agents.
Ingestion points: SKILL.md describes gathering evidence from repository exploration, production logs, Sentry excerpts, and multi-source research.
Boundary markers: The skill encourages isolation through "Assignment Contracts" and the use of fork_turns: "none" to minimize inherited context, though it lacks explicit guidance on sanitizing these inputs for adversarial content.
Capability inventory: The orchestrator utilizes spawn_agent to delegate tasks that include code implementation (Implementation leaf), repository modification, and workflow execution.
Sanitization: No specific mechanisms are defined for sanitizing or escaping content from logs or external research before it is used to define sub-agent objectives.
[COMMAND_EXECUTION]: The skill defines personas and workflows that involve the execution and monitoring of shell commands and CI processes.
Evidence: The Workflow operator persona is explicitly tasked with running "user-authorized workflows" and monitoring "deterministic exports."
Evidence: The Evidence scout persona is instructed to return "commands, timestamps, and exact failures" derived from system health checks and log excerpts.

orchestrate