The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documents installation procedures for several coding agents (Meta Muse Code, Devin, and Google Antigravity) that involve downloading and executing setup scripts from official vendor domains. These are standard procedures for the respective tools and are performed via piped shell commands.
[DYNAMIC_EXECUTION]: The skill uses inline Python scripts, executed via terminal heredocs, to perform utility tasks such as parsing JSONL logs and querying local SQLite databases (e.g., Muse and Devin session indexes). These scripts follow safe practices, such as using parameterized SQL queries, to retrieve session metadata for recovery and supervision.
[COMMAND_EXECUTION]: The skill functions as a router and supervisor for external CLI agents, requiring extensive use of terminal and process tools to drive tools like grok, muse, devin, agy, and codex. This execution is governed by a detailed supervision layer that mandates process isolation and activity monitoring.
[INDIRECT_PROMPT_INJECTION]: As a framework that ingests repository data and external agent outputs into the primary agent's context, the skill manages a significant attack surface for indirect prompt injection. This risk is acknowledged and addressed through instructions for human/parent verification of all artifacts, test-based validation (proving tests can fail), and the use of isolated git worktrees to prevent unauthorized modifications to the production environment.

hermes-delegation