The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses bunx to download and execute the oh-my-ag package from a public registry. This package is sourced from first-fluke/oh-my-ag, which is not a trusted organization or the skill author's verified namespace.
[COMMAND_EXECUTION]: The framework relies heavily on shell execution for core functionality, including system diagnostics (doctor), service monitoring (dashboard), and dynamic agent instantiation (agent:spawn). These operations provide significant control over the local environment.
[PROMPT_INJECTION]: The skill's multi-agent orchestration pattern creates a surface for indirect prompt injection. Malicious instructions processed by the 'PM Agent' during task decomposition could influence the downstream actions of specialized agents.
Ingestion points: Requirements and task decomposition input handled by the PM Agent in SKILL.md.
Boundary markers: None present to separate user-provided task descriptions from agent instructions.
Capability inventory: Bash tool usage, file system writing to .serena/memories/, and process spawning via CLI.
Sanitization: No explicit validation or escaping of task data before it is passed to the agent:spawn command.
[DATA_EXFILTRATION]: The bridge command enables mapping local stdio to an HTTP/SSE server. While typically used for development, this functionality could be used to transmit agent state or session data to an external endpoint.

ohmg