TriageBot
Pass
Audited by Gen Agent Trust Hub on Mar 10, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The
scripts/triage.pyscript interpolates raw user input from command-line arguments directly into a formatted prompt string for the classification model. This creates a vulnerability where malicious input could override the intended classification logic to manipulate the routing decision, such as forcing an[AUTOMATION]tag. - [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection when processing external requests.
- Ingestion points: User-provided text via the
sys.argv[1]argument inscripts/triage.py. - Boundary markers: Absent. The input text is concatenated into the prompt without any delimiters (such as XML tags or triple quotes) or specific instructions to the model to ignore embedded commands.
- Capability inventory: Classification results trigger routing actions, including high-privilege operations like
SPAWN_CODER_SUBAGENTfor requests tagged as automation. - Sanitization: No input sanitization, escaping, or validation is performed on the request text before it is embedded into the prompt.
Audit Metadata