The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it retrieves and processes content (titles, questions, and responses) from an external issue tracker. An attacker with access to the tracker could modify ticket content to influence the agent's behavior during a session.\n
Ingestion points: The SKILL.md file (section 'Parcourir la carte') specifies that the agent loads the map and the full body of tickets from the tracker.\n
Boundary markers: There are no instructions for using delimiters or boundary markers to distinguish untrusted tracker data from system instructions.\n
Capability inventory: The skill allows the agent to execute shell-like commands to invoke other skills (/research, /prototype, /grilling, /domain-modeling), create git branches (research/<name>), and perform network operations via the tracker's API (closing issues, posting comments).\n
Sanitization: The instructions do not mention any validation, filtering, or escaping of the content ingested from the tracker.\n- [COMMAND_EXECUTION]: The skill workflow involves executing several internal tools and sub-agents, such as /setup-mabza-skills for initial configuration and /research for parallel automated investigation.\n- [DATA_EXFILTRATION]: The instructions for resolving 'Tâche' (Task) tickets advise the agent to record the 'location of credentials' (emplacement des informations d'identification) in the ticket response. Documenting sensitive environment details in a shared tracking tool increases the risk of unauthorized data exposure.

wayfinder