The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external task definitions from map.md and the issues/ directory, which could potentially contain malicious instructions.\n
Ingestion points: Step 1 in SKILL.md reads the content of child issues and the project map.\n
Boundary markers: The protocol in references/agent-team-protocol.md specifies 'Context Hygiene' where dispatch prompts are stripped of full issue content, only providing IDs and summarized goals.\n
Capability inventory: The skill uses TeamCreate to spawn sub-agents and manages tracker file updates.\n
Sanitization: The skill implements 'dispatch prompt slimming' (Step 3), ensuring sub-agents receive minimal context, thereby reducing the surface area for instruction injection.\n- [COMMAND_EXECUTION]: The evaluation scripts included with the skill use Python's subprocess module to run internal validation tools.\n
Evidence: evals/test_wayfinder_team_protocol.py executes the check-wayfinder-team-protocol.py script using subprocess.run.\n
Context: This behavior is limited to the test environment and targets local files within the skill package for policy enforcement.\n- [EXTERNAL_DOWNLOADS]: The documentation references an external resource for functional dependency.\n
Evidence: SKILL.md notes that the /wayfinder skill from the mattpocock/skills repository is required.\n
Context: This is a standard functional requirement provided as documentation to the user and does not involve automated remote code execution by the skill itself.

wayfinder-team