The Agent Skills Directory

[COMMAND_EXECUTION]: The /wizard command generates interactive bash scripts intended for infrastructure provisioning and secret configuration.
Evidence: SKILL.md contains the instruction: "Il génère un script bash interactif qui ouvre chaque URL, capture chaque valeur et l'écrit dans les secrets .env et GitHub".
[EXTERNAL_DOWNLOADS]: The /research command performs network operations by delegating reading tasks to background agents that fetch information from external primary sources.
Evidence: SKILL.md states: "délègue les tâches de lecture à un agent d'arrière-plan : il étudie une question par rapport aux sources primaires".
[CREDENTIALS_UNSAFE]: The skill manages the lifecycle of sensitive credentials (tokens, secrets) by capturing them during interactive sessions and persisting them to local .env files and remote GitHub secrets.
Evidence: SKILL.md identifies the capability to "capture chaque valeur et l'écrit dans les secrets .env et GitHub".
[PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection due to its role in processing untrusted external data.
Ingestion points: The skill ingests untrusted data from external sources via the /research tool and from user-provided bug reports or feature requests through the /triage workflow in SKILL.md.
Boundary markers: There is an absence of explicit boundary markers, delimiters, or system-level instructions to ignore embedded commands within the ingested content.
Capability inventory: The agent has the capability to generate and execute shell scripts (/wizard), write persistent files (ADRs, CONTEXT.md, and environment files), and access the network for information retrieval.
Sanitization: No sanitization, validation, or filtering of the external research data or triage inputs is described before the information is processed by the agent.

ask-matt