The Agent Skills Directory

[COMMAND_EXECUTION]: The skill implements a persistence mechanism in Step 1 by instructing the creation of a cron job using CronCreate("execute-task-heartbeat", "* * * * *", "echo heartbeat"). While used as a heartbeat, the use of cron allows for scheduled, recurring execution of commands.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute various local scripts and commands, including python3 script.py, git rev-parse, and .atlas-ai/ship-check.py. This relies on the security of the local project scripts.
[PROMPT_INJECTION]: The skill is explicitly described as 'autonomous by design,' instructing the agent to avoid AskUserQuestion and modal prompts inside the execution loop. This instruction bypasses standard human-in-the-loop safety checkpoints.
[PROMPT_INJECTION]: The skill exhibits an Indirect Prompt Injection attack surface by ingesting and processing external data.
Ingestion points: The skill reads task definitions from .taskmaster/tasks/tasks.json and implementation plans from docs/superpowers/plans/*.md.
Boundary markers: The instructions mandate serializing the full task text into subagent dispatch prompts but do not specify the use of delimiters or 'ignore embedded instructions' warnings.
Capability inventory: The skill has access to Bash (shell execution), Agent (subagent dispatching), Write (file modification), and various MCP tools for project management.
Sanitization: There is no evidence of sanitization, escaping, or validation of the task/plan content before it is interpolated into agent prompts.

execute-task