private-context-bootstrap

Pass

Audited by Gen Agent Trust Hub on May 28, 2026

Risk Level: SAFEDATA_EXFILTRATIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill's primary function is to handle and materialize sensitive user context, including personal profiles, employment records, and identity documents (PDFs). It targets sensitive local directories such as ~/.config/AgentDesk/private and ~/.local/share/AgentDesk/private-repo.
  • [EXTERNAL_DOWNLOADS]: The bootstrap workflow requires cloning or syncing data from remote sources defined by the AGENTDESK_PRIVATE_REMOTE environment variable, as well as downloading PDF documents from remote storage.
  • [COMMAND_EXECUTION]: The instructions explicitly direct the agent to run an external script, scripts/bootstrap-private-context.sh, referenced as part of the OpportunityOS repository.
  • [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing data from remote sources.
  • Ingestion points: The agent reads and processes application-profile.yaml, document-manifest.yaml, field-answer-policy.yaml, personal-profile.yaml, and application-log.jsonl (referenced in SKILL.md).
  • Boundary markers: No explicit boundary markers or instructions are provided to the agent to ignore embedded commands within the ingested data files.
  • Capability inventory: The skill has capabilities for file system access, network operations (git clone/sync), executing decryption tools (age, sops), and running the bootstrap-private-context.sh script.
  • Sanitization: The skill recommends schema validation but does not specify protection against natural language instruction injection within those schemas.
Audit Metadata
Risk Level
SAFE
Analyzed
May 28, 2026, 06:50 PM
Security Audit — agent-trust-hub — private-context-bootstrap