The Agent Skills Directory

[SAFE]: The skill consists entirely of markdown documentation and YAML configuration defining a reasoning workflow. There is no code, shell script, or binary payload included in the package.
[DATA_EXFILTRATION]: The instructions demonstrate security awareness by explicitly warning the agent that fresh review contexts are "never an excuse to leak data or broaden tool authority." This acts as a safeguard against accidental data exposure during isolated analysis passes.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze architectural plans and migration claims (untrusted data). It mitigates the risk of adversarial content within these packets by restricting tool interactions to "safe read-only checks" (e.g., dependency queries, row counts) and explicitly prohibiting the mutation of production environments without separate authority.

decision-challenge