The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell-based commands using tools like codex and claude. Step 4 explicitly suggests a codex exec shell command that processes the contents of the .tasks/ directory, which may contain unsanitized data from repository scans.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill ingests untrusted external data from prd.md (Step 3A) and performs comprehensive scans of existing codebases (Step 3B).
Boundary markers: There are no instructions provided to use boundary markers or delimiters when processing this external content, nor are there instructions to ignore instructions embedded within the ingested data.
Capability inventory: The skill has capabilities to write to the file system, execute shell commands (via codex and claude), and spawn parallel subagents.
Sanitization: The skill lacks explicit sanitization or validation steps for the content derived from the PRD or existing codebase before it is used to influence the agent's planning or tool execution.
[DYNAMIC_EXECUTION]: The skill references a local script ./scripts/shared-init-mb.js to automate file creation. Furthermore, Step 1.2.1 instructs the agent to dynamically generate 'proxy skills' (new SKILL.md files) at runtime to enable native command execution within the agent's environment.

cold-start