krt-state-archivist

Pass

Audited by Gen Agent Trust Hub on Oct 20, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes orchestration state documents which function as a data ingestion surface for the agent. These files may contain untrusted content from various project sources that could influence agent behavior during compaction.
  • Ingestion points: Orchestration state files like docs/orchestration/compound-master-state.md identified in the workflow.
  • Boundary markers: Explicit instructions to load references/safety.md and references/state-contract.md provide guardrails and checklists to separate live context from historical data.
  • Capability inventory: Execution of compact_compound_state.py for file system operations and subsequent manual curation by the agent.
  • Sanitization: A mandatory manual review phase (Step 3) ensures the scaffolded state is accurate, and the compaction script implements logic to halt on ambiguous signals unless explicitly overridden.
  • [COMMAND_EXECUTION]: The skill executes a locally provided Python utility script to perform structural analysis and archiving of state files.
  • Evidence: The workflow in SKILL.md invokes python3 skills/krt-state-archivist/scripts/compact_compound_state.py to perform the archival task.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 20, 2026, 06:43 AM