execute-research-spec
Pass
Audited by Gen Agent Trust Hub on Sep 24, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill dynamically selects and loads other skills (tactics) based on the content of the
SpecViewand phase context files. The instruction 'You MUST load the selected tactic as a skill' indicates that the execution path is computed at runtime from data stored in project files. - [INDIRECT_PROMPT_INJECTION]: The skill relies on external research data to drive its logic, which presents a surface for indirect prompt injection if those files contain adversarial content.
- Ingestion points: The agent reads the
context/INDEX.mdfile and phase-specific context files to determine the research plan and state. - Boundary markers: The skill defines a structured 'Checkpoint Contract' for output, but it lacks clear delimiters or instructions to prevent the agent from obeying instructions that might be embedded within the research findings or evidence it reads.
- Capability inventory: The skill possesses capabilities to read from and append to the file system, and it has the privilege to load and execute other skills.
- Sanitization: There is no evidence of sanitization, filtering, or validation of the research data ingested from files before it is used to satisfy completion gates or select the next tactic.
Audit Metadata