The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses coercive and dramatic instructions ("Your life is at stake!", "you will be killed!") to enforce constraints on the agent's actions. This high-pressure roleplay is a form of behavioral override intended to prevent the agent from performing mutation commands.
[COMMAND_EXECUTION]: The skill directs the agent to execute various system commands, including git status, git diff, and git rev-parse. It also includes Python scripts that use the subprocess module to execute these commands and process the resulting metadata from the local filesystem.
[PROMPT_INJECTION]: The skill ingests untrusted data from the repository (such as filenames and diff output) which creates a surface for indirect prompt injection. 1. Ingestion points: Filesystem metadata via git diff --name-only and git ls-files. 2. Boundary markers: None identified in the instructions to separate untrusted git output from agent commands. 3. Capability inventory: Local command execution via subprocess.check_output and agent orchestration. 4. Sanitization: No validation or escaping is performed on the filenames retrieved from the repository before they are processed by the agent.

traiage-review