ktl-prose

Pass

Audited by Gen Agent Trust Hub on Oct 12, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes shell commands including git and project-specific shell scripts (knowledge-preflight.sh, knowledge-conventions.sh). Subprocess calls in the bundled Python script are used to retrieve file versions from git, with input validation performed on revision references to prevent injection.
  • [DYNAMIC_EXECUTION]: The agent is instructed to execute a local Python script (scripts/prose-check.py) to perform complex prose analysis and verify that no functional changes occurred during rewording. The script is part of the skill's local distribution.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes Markdown files that may contain untrusted content. This risk is mitigated by explicit instructions to treat the text as data/material rather than instructions, and by the use of a verification script that ensures links, code, and numbers remain unchanged.
  • Ingestion points: Markdown files in .lokf/knowledge/ or user-specified paths (e.g., references/other-files.md).
  • Boundary markers: Explicit instructions in SKILL.md to treat reworded text as material, not instructions; automated verification script reporting.
  • Capability inventory: File read/write, local command execution (bash, python3, git, just).
  • Sanitization: Automated detection of non-printing/unseen characters (e.g., zero-width spaces, Trojan Source vectors) and verification of structural consistency (link targets, code blocks).
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 12, 2026, 12:23 AM