maintain-workflow-context-restore

Pass

Audited by Gen Agent Trust Hub on Jun 23, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: Executes a local shell command ls .claude/checkpoints/*.md | sort to retrieve and order available checkpoint files by their timestamped filenames.- [PROMPT_INJECTION]: The skill processes content from external checkpoint files which represents an indirect prompt injection surface as these files could contain malicious instructions.
  • Ingestion points: Reads and parses Markdown files with YAML frontmatter from the local .claude/checkpoints/ directory.
  • Boundary markers: Relies on YAML keys and Markdown headers to structure the data, though it does not explicitly instruct the agent to ignore instructions embedded within the text fields.
  • Capability inventory: The skill's own operations are explicitly restricted to read-only file access; however, it recommends a 'Next Command' to the user based on file contents.
  • Sanitization: The skill extracts specific metadata fields but does not sanitize the text values within those fields before displaying them or using them for context.
  • Mitigation: Includes a mandatory 'Hard Gate' that prohibits the agent from automatically writing code or executing commands without presenting a summary and receiving explicit user confirmation, which serves as a robust defense against autonomous obedience to injected instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 23, 2026, 05:15 AM
Security Audit — agent-trust-hub — maintain-workflow-context-restore