The Agent Skills Directory

Execution of reporter code: The skill's primary function involves extracting and executing code provided by external issue reporters. This is performed by calling a specialized reproduction skill and represents a pattern of executing untrusted external content. To mitigate the risk of executing potentially harmful code, the skill mandates a verified isolated execution environment.
Indirect Prompt Injection surface: The skill ingests untrusted data from external issue trackers such as GitHub or Jira.
Ingestion points: Issue descriptions and comments are retrieved during the pool selection process and during per-issue triage.
Boundary markers: The skill includes explicit instructions for the agent to treat external content as data and never as instructions, though it does not describe specific structural delimiters for data processing.
Capability inventory: The skill possesses the capability to execute code (via sub-skills), write data to disk (verdict.json), and perform network operations to communicate with issue trackers.
Sanitization: The skill relies on classification logic and agent instructions rather than automated sanitization of the external text content.
Local command execution: The skill performs several local operations, including the execution of Python modules (e.g., setup_preflight) and version control commands (e.g., git status). These commands are used to verify the project configuration and ensure the working environment is clean.
External service interaction: The skill performs network requests to interact with issue tracking platforms like GitHub and Jira to fetch data. These operations are essential for the skill's triage and reassessment functionality.

reassess