knowhow-capture

Pass

Audited by Gen Agent Trust Hub on Jun 18, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes a basic shell command (ls -d .knowhow/) to verify the existence of the workspace repository. This is a standard and safe operational check.
  • [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection as it processes external, untrusted content from chat logs and user-provided files. 1. Ingestion points: Reads conversation history and external file content (SKILL.md, Step 1). 2. Boundary markers: No specific delimiters are defined for source data. 3. Capability inventory: Utilizes file read/write operations and basic shell commands. 4. Sanitization: Every extracted item must be manually approved, edited, or rejected by the user (Step 3) before it is committed to the inbox, serving as a primary security checkpoint.
  • [SAFE]: All file operations are restricted to the local workspace and follow a well-documented structure described in the reference files. No network operations or external data exfiltration patterns were detected.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 18, 2026, 07:02 AM
Security Audit — agent-trust-hub — knowhow-capture