The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill generates a handoff prompt for a successor agent by ingesting potentially untrusted data from the repository environment.
Ingestion points: The skill reads project files, version control diffs, build logs, and plans from the workspace (SKILL.md).
Boundary markers: The skill recommends using verified, inferred, and unknown labels to demarcate the trust levels of summarized information (SKILL.md).
Capability inventory: The generated output template (assets/handoff.template.md) instructs the successor agent to execute shell commands and perform file operations based on the handoff content.
Sanitization: The skill explicitly instructs the agent to remove secrets, tokens, and credentials. It does not provide specific sanitization for natural language instructions embedded in ingested data, but emphasizes verification against inspectable evidence.
[SAFE]: The skill references documentation and guidance from well-known and trusted technology organizations including Anthropic, OpenAI, Microsoft, and the LangChain project.
Evidence: references/source-notes.md contains links to official documentation on anthropic.com, openai.github.io, docs.langchain.com, and learn.microsoft.com.

session-handoff