The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and audit the entire contents of a Git repository, which constitutes untrusted third-party data.
Ingestion points: The agent reads all tracked and untracked files within the Git repository to build a coverage ledger and perform analysis (SKILL.md, scripts/sweep-ledger.py).
Boundary markers: There are no explicit instructions or delimiters defined to prevent the agent or its subagents from obeying instructions embedded within the source files or documentation being audited.
Capability inventory: The skill is authorized to perform automated fixes, run repository checks (lint, test, build), and spawn subagents (using opus and sonnet models) to modify the codebase (SKILL.md).
Sanitization: The skill lacks explicit sanitization logic to handle potentially adversarial content within the code it processes.
[COMMAND_EXECUTION]: The skill utilizes a local Python script to manage repository state by executing Git commands.
Evidence: scripts/sweep-ledger.py uses subprocess.run to call git status, git ls-files, and other repository management commands. While these calls use list-based argument passing to mitigate shell injection risks, they provide the skill with comprehensive visibility into the local repository structure and state.

fresh-eyes-sweep