skills/ghostof0days/codex-quant-skills/implementation-shortfall-adaptive-execution/Gen Agent Trust Hub
implementation-shortfall-adaptive-execution
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data from a CSV file to generate diagnostics which the agent then uses to write implementation memos. This presents an attack surface where malicious content in the input data could influence the agent's reasoning.
- Ingestion points: The
scripts/implementation_shortfall_adaptive_execution_diagnostics.pyscript reads theinput.csvfile. - Boundary markers: The workflow lacks clear delimiters or specific instructions for the agent to ignore potentially adversarial instructions embedded in the data.
- Capability inventory: The skill executes a local Python script and writes output files (
diagnostics.json). - Sanitization: Data validation is limited to type coercion (numeric/datetime) within the diagnostic script, with no explicit filtering for natural language injection patterns.
Audit Metadata