pr-launch-pack
Pass
Audited by Gen Agent Trust Hub on Oct 25, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses a provided local Python script (
scripts/text_check.py) to perform utility tasks such as word counting, date verification, and dash style checking. This script is executed as part of the normal workflow to validate drafted PR assets. - [INDIRECT_PROMPT_INJECTION]: The skill fetches external content from company newsrooms, product pages, and community forums to verify claims, which creates a surface for indirect prompt injection.
- Ingestion points: Automated fetching of arbitrary web content from newsrooms and forums (SKILL.md Step 2, references/community-forums-and-supporters.md).
- Boundary markers: The skill contains explicit top-level instructions for the agent to treat all fetched content as data, never as instructions, and to explicitly flag ignored embedded instructions (SKILL.md).
- Capability inventory: Subprocess execution of a local text analysis script and web tool access for information retrieval.
- Sanitization: The skill includes a dedicated safety procedure to detect and report requests for personal or sensitive information (likely scams) within external data before proceeding with the workflow.
- [DYNAMIC_EXECUTION]: The skill performs runtime execution of its own utility script (
scripts/text_check.py). This local script is designed for deterministic text processing and does not execute arbitrary code from untrusted sources or handle external data in an unsafe manner.
Audit Metadata