The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to read local and user-level configuration files (.claude/settings.json, ~/.claude/settings.json, and .claude/settings.local.json) using shell commands (cat and grep). These files are part of the platform's core configuration and may contain sensitive data; the skill reads them to verify if certain workflow hooks are registered.
[PROMPT_INJECTION]: The skill contains "CRITICAL CONSTRAINTS" that explicitly forbid the agent from using platform-native tools like EnterPlanMode and ExitPlanMode, overriding the agent's default behavior and completion flow to enforce its own logic.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (a project specification or requirements document) to generate plans and tasks, creating an attack surface where instructions embedded in the source data could influence the agent's actions.
Ingestion points: The input specification or requirements document referenced in the Overview and Scope Check sections of SKILL.md.
Boundary markers: The plan template in SKILL.md includes a **Spec:** header to track the source file, but the skill lacks specific instructions to sanitize the spec's content or warn the agent against executing instructions found within it.
Capability inventory: As defined in SKILL.md, the skill uses TaskCreate and TaskUpdate (which include executable verifyCommand metadata fields), shell commands via cat and grep, and file system writes to the docs/superpowers/plans/ directory.
Sanitization: No validation, filtering, or escaping of content extracted from the untrusted specification document is specified.
[DYNAMIC_EXECUTION]: The skill generates and stores executable shell commands (such as pytest, bash, and git commands) within task descriptions and JSON metadata, which are designed to be executed by the agent or its subagents during the plan implementation phase.
[COMMAND_EXECUTION]: The skill directs the agent to invoke other skills (superpowers-extended-cc:subagent-driven-development, superpowers-extended-cc:executing-plans, and superpowers-extended-cc:using-git-worktrees) and utilizes core agent tools such as TaskList, TaskCreate, and AskUserQuestion to manage the planning and execution workflow.

writing-plans