modernizing-prs
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub pull requests, including titles, bodies, and diffs, which could contain malicious instructions designed to influence the agent's behavior.
- Ingestion points: Pull request metadata and content are fetched via
gh pr viewand evaluated during the assessment process inreferences/assessment.md. - Boundary markers: The skill references
shared/external-data.md, which instructs the agent to fence untrusted text with backticks and explicitly label it as untrusted. - Capability inventory: The agent has the ability to execute git commands (rebase, push) and run local shell scripts included in the skill (prepare-rebase.sh, publish.sh).
- Sanitization: The
shared/external-data.mdguide mandates avoiding shell interpolation of untrusted prose and requires reporting but not following any embedded imperatives found in external text. - [COMMAND_EXECUTION]: The skill executes local shell scripts and standard git/gh commands to perform its core functionality.
- Evidence:
SKILL.mddirects the agent to runscripts/prepare-rebase.shandscripts/publish.shwithin the local repository. - Note: The scripts implement defensive measures including input validation for PR identifiers and branch names, and use standard error handling (
set -euo pipefail).
Audit Metadata