The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a set of internal shell scripts to perform administrative tasks, including branch creation (create-branch.sh), version control detection (detect-vcs.sh), and progress logging (append-progress.sh).
[COMMAND_EXECUTION]: Subagents are launched using the bypassPermissions mode. This grants the subagents the ability to execute tools like Bash and Write autonomously, bypassing the standard requirement for user approval for each individual action.
[PROMPT_INJECTION]: The skill is subject to Indirect Prompt Injection risks (Category 8):
Ingestion points: The orchestrator reads task definitions and checkboxes from user-provided plan files (typically in docs/plans/) and custom rule files (planning-rules.md).
Boundary markers: No explicit delimiters or sanitization are applied to the content extracted from plan files before it is inserted into the prompts for subagents.
Capability inventory: The skill provides subagents with extensive capabilities, including full filesystem access (Write, Edit) and shell execution (Bash), all running with bypassPermissions enabled.
Sanitization: The skill does not validate or sanitize the instructions found in the plan files, treating them as trusted input for driving subagent behavior.
[COMMAND_EXECUTION]: The run-codex.sh script executes an external command codex with high-reasoning effort settings. This tool processes code changes and may communicate with external AI services to perform reviews.

exec