The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local git and gh (GitHub CLI) commands for repository management and branch cleanup. All destructive actions, such as branch deletions, are gated by a safety model requiring per-item user confirmation and 'force' flags only after seeing unmerged commits.
[EXTERNAL_DOWNLOADS]: In the 'Publications Researcher' step, the skill can perform external searches using tools like WebSearch and WebFetch. This functionality is explicitly disabled by default and requires the agent to pause and ask the user for permission before any network communication occurs.
[PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection because it reads and processes the target repository's source code and external web search results to generate documentation and reports. This is a low-severity surface inherent to its function; the skill mitigates this by showing all proposed diffs and markdown reports to the user for verification before any files are written or updated.
[DATA_EXFILTRATION]: The skill transmits specific project metadata (e.g., project name and primary authors) to external search engines during the optional publication search phase. This transmission is limited, purposeful, and requires user opt-in.

oss-ready-flow