The Agent Skills Directory

[SAFE]: The skill provides a robust safety procedure for human-in-the-loop validation of dangerous operations. It emphasizes least privilege, explicit user confirmation for irreversible actions, and mandatory state-checking before writes.
[NO_CODE]: The skill consists entirely of markdown instructions and YAML metadata; it performs no automated actions and does not ship with any scripts, binaries, or configuration files.
[INDIRECT_PROMPT_INJECTION]: While the skill processes information about 'mutations' which could theoretically be attacker-influenced, it actively mitigates this risk by requiring 'stable IDs' and prohibiting referential pronouns like 'it' or 'the above,' forcing the agent to explicitly re-verify every target.

preflight-mutations