The Agent Skills Directory

[SAFE]: The skill implements a sophisticated multi-wave exploration pattern using the Agent tool to dispatch parallel Explore subagents. This is a legitimate architectural pattern for complex codebase research.
[SAFE]: Data processing is focused on evidence-backed synthesis (file:line references). The references/synthesis.md file defines a clear algorithm for normalizing findings, deduplicating claims, and calculating completeness based on novelty rates and open questions.
[SAFE]: The frontmatter correctly scopes tool usage to Read, Grep, Glob, Bash, and Agent. Analysis of the instructions shows these tools are used for their intended purposes (e.g., Grep for scope estimation, Agent for sub-tasks).
[SAFE]: The skill includes explicit security and privacy guardrails, such as evidence redaction rules and a strict verification checklist that forbids git add/commit/push operations.
[SAFE]: Information passing between waves is restricted to facts and open questions, explicitly instructing agents not to treat previous conclusions as absolute truth, which reduces the risk of feedback loops or error propagation.

deep-explore