wiki-playbook-review

Pass

Audited by Gen Agent Trust Hub on Nov 17, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the Obsidian vault (meeting notes, journals, postmortems) which could contain malicious instructions designed to alter the playbook in unauthorized ways. The skill mitigates this through two primary controls:
  • Explicit Instruction: It includes a specific warning to the agent that evidence pages are untrusted and that user quotes should be treated as data points rather than direct instructions.
  • Mandatory Staging: The skill is architected to never write to live files; instead, it writes to a _staging/ directory and requires the human owner to review the diff and run a separate command (/wiki-stage-commit) to apply changes.
  • [COMMAND_EXECUTION]: The skill uses the obsidian-wiki CLI tool for vault querying and logging. These are local operations scoped to the user's defined vault path and do not involve remote execution or network exfiltration.
  • [SAFE]: The skill demonstrates best practices for agent safety, including state management (preventing duplicate reviews), clear boundary markers for untrusted data, and a 'human-in-the-loop' requirement for all write operations.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 17, 2026, 05:31 PM