The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The orchestration instructions in SKILL.md (Step 0) advise running PowerShell with the -ExecutionPolicy Bypass flag. This is used to circumvent local execution policies that would otherwise prevent the script from running, which is a common but sensitive privilege escalation pattern.
[COMMAND_EXECUTION]: The create_github_issues.py script executes shell commands using subprocess.run to retrieve authentication tokens from the GitHub CLI (gh auth token). Additionally, SKILL.md triggers local PowerShell and Python scripts to perform environment checks and issue creation.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted data from various external sources (e.g., 'ANY input', 'audit spreadsheet', 'QA report'). Maliciously crafted content in these files could attempt to influence the agent's behavior during the extraction or classification phases.
Ingestion points: Findings are extracted from user-provided documents like Excel, CSV, or meeting notes as described in SKILL.md Step 1.
Boundary markers: The skill lacks technical delimiters for the ingested content, relying instead on 'Safety Gates' involving human review of a visual dry-run.
Capability inventory: The skill has the ability to read local files, execute subprocesses for authentication, and perform network requests to the GitHub API to create labels, milestones, and issues.
Sanitization: The Python script create_github_issues.py maps data directly from the input JSON to GitHub Issue fields (title, body, labels) without explicit sanitization or filtering logic.

findings-to-github-issues