The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell commands to gather project metadata and history.
Evidence: Executes git status, git log --since="1 week ago", and git diff to extract commit information and code changes.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple sources and interpolates it into prompts for auxiliary agents without explicit sanitization.
Ingestion points: Data enters via git log (commit messages), git diff (code changes), local files (results, analysis scripts, configurations), and external WebSearch results for paper verification.
Boundary markers: The instructions use square-bracketed placeholders like [Include summary of changes] and [Include the draft report] when calling the report-planner and report-reviewer agents, but they lack explicit warnings to the model to ignore instructions embedded within that data.
Capability inventory: The skill possesses capabilities for shell command execution (git), file system access (read and write), external network access (WebSearch), and the ability to trigger other agent tasks via the Task tool.
Sanitization: There is no evidence of escaping, filtering, or sanitization of the content extracted from the repository before it is passed to the AI agents, creating a surface where a malicious contributor could influence agent behavior through crafted commit messages or file contents.

weekly-report-writer