github-project-init
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from local markdown files to automate the creation of GitHub issues, milestones, and project boards. This creates a surface for indirect prompt injection if the source documents (e.g., task breakdowns) are untrusted.
- Ingestion points: The skill reads from
docs/task-breakdown/,docs/business/sprint-breakdown.md,docs/technical-specs/, anddocs/deployment-plan/(SKILL.md). - Boundary markers: No explicit delimiters or instructions to ignore embedded commands are used when parsing these files.
- Capability inventory: The skill utilizes the GitHub CLI (
gh) to perform significant administrative actions, including creating repositories, modifying branch protection rules, and bulk issue creation. It also performs file system writes to the.github/directory. - Sanitization: No sanitization or validation of the ingested document content is described prior to its use in CLI commands or file generation.
- [DYNAMIC_EXECUTION]: The skill dynamically generates and commits executable content to the repository.
- It creates
.github/workflows/ci.ymland.github/workflows/build.yml(ordeploy.yml) based on project-specific commands and detected stacks. - It writes a Python script,
.github/scripts/pr_hygiene.py, which is executed during the CI process to validate PR metadata. - [COMMAND_EXECUTION]: The skill executes high-privilege administrative commands using the
ghCLI and the GitHub API. - It modifies sensitive repository settings, such as branch protection rules (
gh api -X PUT ... branches/{b}/protection) and default branch configurations, which require significant repository permissions.
Audit Metadata