The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection because it ingests and processes untrusted data from external GitHub issues and comments.
Ingestion points: The scripts/inspect_issue.py script fetches issue metadata, bodies, and comments via the GitHub API (gh issue view and gh api).
Boundary markers: While SKILL.md specifies a structured Issue Analysis Report format, there are no explicit instructions or delimiters used to tell the agent to ignore potentially malicious instructions embedded within the fetched GitHub content.
Capability inventory: The agent has the capability to search the codebase, propose fix plans, modify files, and commit/push changes via Git (documented in SKILL.md). The helper script scripts/inspect_issue.py executes shell commands using subprocess.run.
Sanitization: The script extracts specific patterns (errors, stack traces) using regular expressions but does not perform sanitization or filtering of the extracted text before it is presented to the agent for analysis.
[COMMAND_EXECUTION]: The bundled script scripts/inspect_issue.py executes external shell commands (gh and git) using the subprocess module. These calls are used for the skill's legitimate functions (fetching data and checking file existence), but they represent a surface for potential misuse if the arguments passed to these commands were not properly resolved.

gh-fix-issue