retro
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill contains instructions to parse and execute "instruction blocks" found in the output of the
gstack-skill-startutility. - Ingestion points: Standard output from the
gstack-skill-starttool call. - Boundary markers: Uses
GSTACK_INSTRUCTION_BEGINandGSTACK_INSTRUCTION_ENDdelimiters. - Capability inventory: The skill has permissions for
BashandWrite, which could be utilized by these dynamic instructions. - Sanitization: Employs a
SESSION_IDcheck to ensure instructions are only honored if they match the ID generated in the current execution turn. - [INDIRECT_PROMPT_INJECTION]: The skill reads and processes local historical data including prior retrospectives, learning logs, and timeline events to generate its narrative.
- Ingestion points: Files matching
*.jsonand*.jsonlpatterns within the project's state root. - Boundary markers: Absent; data is read and interpolated directly.
- Capability inventory:
Bash,Read,Write. - Sanitization: None; the content is treated as trusted historical context.
- [COMMAND_EXECUTION]: The skill performs extensive execution of local binaries and scripts for metrics collection and telemetry.
- Evidence: Calls to
gstack-retro-metrics,gstack-global-discover, andgstack-factory-scorecardlocated in the vendor's binary directory. - Context: These tools are part of the vendor's specific framework and are executed from the user's home directory or local project paths.
Audit Metadata