Writing Plans
PassAudited by VirusTotal on May 12, 2026.
Overview
Type: OpenClaw Skill Name: writing-plans Version: 0.1.0 The skill bundle is designed to instruct an AI agent on how to generate detailed implementation plans in markdown format. The `SKILL.md` file contains instructions for structuring the plan, including example code snippets and shell commands (e.g., `pytest`, `git`) that are intended to be *written into the plan*, not executed by the agent as part of this skill. The agent is instructed to save the generated plan to a local file path (`docs/plans/`), which is a benign operation. There are no instructions for data exfiltration, malicious execution, persistence, obfuscation, or prompt injection beyond guiding the agent's output formatting and suggesting legitimate follow-up skills (`superpowers:executing-plans`, `superpowers:subagent-driven-development`).
Findings (0)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
A future agent may follow the saved plan’s instructions, so mistakes or overly broad steps in the plan could carry into implementation.
The skill intentionally writes persistent plan files that include instructions aimed at a future Claude session. This is transparent and aligned with the planning workflow, but persisted instructions can influence later agent behavior.
**Save plans to:** `docs/plans/YYYY-MM-DD-<feature-name>.md` ... `> **For Claude:** REQUIRED SUB-SKILL: Use superpowers:executing-plans to implement this plan task-by-task.`
Review generated plan files before using them to drive execution, especially the listed files, commands, and handoff instructions.
If the user chooses subagent-driven execution, separate agents may act on portions of the plan, making review checkpoints important.
The skill describes an optional workflow where tasks are handed to fresh subagents. This is disclosed and user-selected, but it introduces inter-agent handoff that users should supervise.
**1. Subagent-Driven (this session)** - I dispatch fresh subagent per task, review between tasks, fast iteration
Use the subagent option only when comfortable with task-by-task delegation, and keep the stated review-between-tasks checkpoint.