The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill orchestrates a multi-agent flow where it ingests and processes data from worker subagents (Executor, QA, and Fix workers) that interact with external project files.
Ingestion points: Worker final reports, executor return contracts, and QA certification results stored in the KYRO FLOW STATE (SKILL.md).
Boundary markers: The skill uses structured Markdown templates (e.g., ## STATUS, ## FINDINGS) for contract reporting, but does not implement explicit boundary delimiters or instructions to ignore instructions embedded within the worker's output.
Capability inventory: The orchestrator can execute shell commands via Bash, read local files, and launch subagents with dynamic model configurations (SKILL.md).
Sanitization: The instructions mandate verbatim presentation of worker outputs ("Never paraphrase", "present the worker's final message verbatim") and incorporate this raw data into the generation of "Fix Plans," which could allow malicious instructions in processed files to influence orchestrator or user actions.
[COMMAND_EXECUTION]: The orchestrator performs legitimate interactions with the local environment to manage the sprint workflow.
It executes the kyro status CLI command to identify the active sprint stage (SKILL.md).
It reads platform-specific configuration files located at ~/.pi/agent/settings.json and ~/.pi/agent/models-store.json to enumerate and recommend available AI models for the QA phase on the Pi platform (SKILL.md).

kyro-pi-sprint-flow