The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data in the form of code artifacts, SQL, and design snippets for review. However, it implements strong boundary markers through its adversarial prompt template ("Assume the author is overconfident", "Do NOT validate", "Do NOT summarize"), which forces the agent to maintain a skeptical persona, effectively mitigating the risk of following instructions embedded within the processed data.
[COMMAND_EXECUTION]: The skill does not contain any shell command execution, file system writes, or system configuration modifications.
[DATA_EXFILTRATION]: There are no network operations (e.g., curl, wget) or hardcoded credentials detected in the skill logic or evaluation files.
[REMOTE_CODE_EXECUTION]: The skill relies on local sub-agent invocation via platform tools rather than downloading or executing remote scripts.

sdlc-doubt