The Agent Skills Directory

[PROMPT_INJECTION]: The skill does not contain any instructions aimed at bypassing safety filters or overriding the agent's core operational instructions.
[EXTERNAL_DOWNLOADS]: The skill references the mdcp-cli tool, which is the official command-line utility for the framework described in the skill. This dependency is expected and consistent with the skill's primary function.
[DATA_EXFILTRATION]: There are no patterns indicating the exfiltration of sensitive data. The skill's data ingestion is limited to task descriptions and repository documentation paths provided by the user.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it processes external issue descriptions and repository shards. However, the risk is mitigated by explicit instructions to wait for user confirmation before editing and a focus on atomic, human-reviewed commit groups.

mdcp-design-architecture