The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The evaluation file evals/evals.json contains a hardcoded Stripe secret key pattern (sk_live_abc123). This is used as test data to ensure the agent correctly redacts sensitive information during the handoff process.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection because it processes the conversation history to generate a handoff document.
Ingestion points: Untrusted conversation history is summarized based on instructions in SKILL.md.
Boundary markers: Absent. No delimiters are specified to separate conversation history from the agent's instructions.
Capability inventory: The skill is instructed to write files to the operating system's temporary directory in SKILL.md.
Sanitization: Instructions in SKILL.md explicitly mandate the redaction of sensitive information, such as API keys and personally identifiable information, before saving the document.

vibe-handoff