The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, specifically code diffs and project implementation plans, which creates a surface for indirect prompt injection.
Ingestion points: According to references/orchestrator-guide.md and references/stage-procedures.md, the agent accepts diff: branch changes and plan: <path or text> as primary runtime inputs.
Boundary markers: The skill lacks explicit instructions to treat the ingested code or plan content as data only or to ignore any embedded instructions (e.g., within code comments).
Capability inventory: The skill assumes the ability to launch and resume subagents (agent_id, resume), read files, and generate markdown reports. No direct shell execution or network access is specified in the provided instructions.
Sanitization: There are no instructions for sanitizing or escaping the content of the diffs or plans before processing.

kf-g-agent-review-post-implementation-two-stage