The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data by reading code changes and surrounding context from the repository. This data acts as a vector for indirect prompt injection, where malicious instructions hidden in code comments or files could influence the agent's behavior during the cleanup process.
Ingestion points: Reading changed files and surrounding code to establish contracts and conventions.
Boundary markers: None specified for separating the ingested code from the agent's instructions.
Capability inventory: Writing file changes, running project-specific formatters/checks, and executing Git commands (status, diff, add, commit).
Sanitization: No explicit sanitization or validation of the processed code content is mentioned.
[COMMAND_EXECUTION]: The instructions require the agent to execute shell-based tools, including project-specific formatters and Git commands. When the --auto flag is used, these commands (including git commit) are executed without requiring manual user confirmation, increasing the impact of any potential automated errors or injections.

baethon-deslop