timeline-report

Warn

Audited by Gen Agent Trust Hub on Oct 28, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill accesses local application files including ~/.claude-mem/settings.json and the ~/.claude-mem/claude-mem.db SQLite database to extract project history and configuration details.
  • [COMMAND_EXECUTION]: Shell commands such as curl, git, and bash are executed to perform project environment detection and interact with the local worker API.
  • [DYNAMIC_EXECUTION]: The skill uses node -e to execute a script that parses local configuration files at runtime.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted timeline observations and passes them to an LLM subagent for analysis, creating a potential surface for indirect prompt injection.
  • Ingestion points: External project observations retrieved from http://localhost:${WORKER_PORT}/api/context/inject in SKILL.md.
  • Boundary markers: The timeline content is placed directly into the agent's prompt without specific delimiters or warnings to ignore embedded instructions.
  • Capability inventory: The subagent is explicitly instructed to execute sqlite3 commands on the ~/.claude-mem/claude-mem.db database.
  • Sanitization: There is no evidence of sanitization or filtering of the project observations before they are analyzed.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 28, 2026, 01:06 AM