blueprint
Pass
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions explicitly direct the agent to execute shell commands defined in the "Verify" field of the generated plan file.
- Evidence:
SKILL.mdspecifies "Run each Verify once (2>&1 | tail -5)". Thereferences/artifact.mdfile defines the schema for these commands, where<command>is provided as a template for execution. - [INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection as it ingests data from the repository to formulate execution steps.
- Ingestion points: The skill reads from the existing repository (
SKILL.mdstep 1) and accepts user input to define "Locked constraints" (referenced inreferences/grill.md). - Boundary markers: The skill does not define specific delimiters or instructions to ignore commands embedded within the repository files it reads.
- Capability inventory: The agent has the capability to execute shell commands (
SKILL.md) and invoke other skills (such as thebuildskill). - Sanitization: There is no evidence of sanitization or validation of the commands generated from the repository facts before they are executed in the verification step.
- [DYNAMIC_EXECUTION]: The skill performs dynamic execution by generating a plan containing shell commands and then immediately executing those commands.
- Evidence: The workflow involves writing a plan to
docs/plans/<YYYY-MM-DD>-<slug>.mdand then running eachVerifycommand once as specified in the second part of theSKILL.mdinstructions.
Audit Metadata