The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted code and diffs from the active repository to generate handoff documentation, creating a surface where malicious instructions in project files could be included in the handoff and processed by a future session.\n
Ingestion points: scripts/git_changes.py reads diff hunks and file content directly from the project repository to provide context for the handoff.\n
Boundary markers: The skill relies on Markdown headers in the references/handoff-template.md to delimit sections but does not use explicit 'ignore instructions' delimiters for the extracted snippets.\n
Capability inventory: scripts/handoff.py writes new files to the filesystem; scripts/git_changes.py executes local git commands.\n
Sanitization: Filenames are sanitized via a slugify function in scripts/handoff.py to prevent path traversal, but the textual content extracted from the repository is included without filtering.

handoff