The Agent Skills Directory

[PROMPT_INJECTION] (MEDIUM): Vulnerability to indirect prompt injection through staged repository content.
Ingestion points: SKILL.md workflow step 2 ingests data via get_changed_files or git diff --cached.
Boundary markers: Absent; the agent is directed to analyze the diff content directly without delimiters or instruction isolation.
Capability inventory: Executes local subprocesses git rev-parse and git diff (SKILL.md); no network or file-write capabilities identified.
Sanitization: Absent; the skill processes raw diff output.
[COMMAND_EXECUTION] (LOW): The skill executes standard Git commands for local repository introspection.
Evidence: git rev-parse --show-toplevel and git diff --cached are used in SKILL.md to resolve context and retrieve data.

commit-message