wayfinder
Pass
Audited by Gen Agent Trust Hub on Nov 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill manages a workflow by reading and writing markdown files (
map.mdand tickets) which are used to determine subsequent actions. \n - Ingestion points: The agent reads
map.mdand individual ticket files intickets/to identify the 'frontier' and instructions for the current session. \n - Boundary markers: The skill uses YAML frontmatter and markdown headers but lacks explicit instructions to the agent to ignore or sanitize potentially malicious instructions embedded within the ticket 'Question' or 'Resolution' sections. \n
- Capability inventory: The skill can spawn subagents (
spawn_agent), perform network-enabled research ($web-research), and write files to the repository. \n - Sanitization: There is no mechanism to validate or sanitize the content of tickets before they are processed by the agent or passed to subagents.\n- [PROMPT_INJECTION]: The skill uses directive language to control the agent's operational environment and file-handling behavior. \n
- The 'Codex compatibility note' contains instructions to 'read that file in full yourself (every line, with no line limit)', which is an attempt to override standard platform limitations on context window management and file reading. \n
- The 'Closing Reminders' section uses the prefix 'IMPORTANT MUST ATTENTION' to enforce strict adherence to the skill's 'plan, do not build' protocol and other operational constraints.\n- [COMMAND_EXECUTION]: The skill orchestrates the use of multiple high-capability tools and sub-agents. \n
- It explicitly claims authorization and triggers the use of
spawn_agentfor research tasks. \n - It integrates with external tools like
$web-research(network access),$ui-design, and$investigateto perform its planning functions.
Audit Metadata