Plan Mode
Write an actionable Markdown plan to .hermes/plans/ — no code execution or file changes.
The skill explicitly restricts to read-only operations: no code execution, no editing project files, no mutating terminal commands, deliverable is only a plan file, demonstrating least privilege. However, user confirmation relies on conversational context inference without explicit ack step, and no sensitive-data or rollback handling is documented, hence deductions.
Instructions are clear with specific save path format and filename generation, but no error handling or failure feedback is provided (e.g., behavior when path unwritable or write fails). Relies on runtime provided path or self-creation, does not specify behavior when out of scope. No tests are visible statically, so per anchor max 10, given 8.
Use scenario is clear (plan mode) with boundaries (no execution, only plan file), and trigger is inferred from conversation context. But no concrete non-fit examples, input/output examples, or environment fit (Chinese support, network reachability) mentioned, thus below full marks.
File structure is clean with metadata (version, author, license), usage and save location, but lacks installation dependencies, changelog, known limitations, or maintenance responsibility. Version exists but no change history, so per anchor range 7-15, given 8.
Static assessment cannot verify actual execution. Goal is reasonable (produce actionable plan) and deliverable is a plan file directly, but no output examples or test cases, marginal benefit not quantified, so per static anchor max 7 is appropriate.
No tests, CI, or reproducible evidence; only source definitions. README provides some architecture claims but not covering this skill's key paths, per static read anchor max 5, but due to zero evidence given 2.
- The skill relies on conversational context inference, which may trigger inappropriately; explicit user confirmation is recommended.
- No error handling or failure feedback is provided; write failures might be silent, requiring enhancement.
- Static review cannot confirm demonstration; test in a controlled environment before deployment.
What does this skill do, and when should you use it?
This skill switches the agent into 'plan mode,' producing only an actionable Markdown plan saved under .hermes/plans/ in the current workspace. It forbids implementing code, editing project files (except the plan file), and running mutating commands; only read-only inspection is allowed. The plan includes goal, context, approach, step-by-step steps, files likely to change, tests, risks, and open questions. If no explicit request, it infers the task from conversation context or asks a clarifying question. After saving, the agent briefly replies with the plan summary and path.
Restricts the agent to planning only. It writes a Markdown plan using the write_file tool to .hermes/plans/YYYY-MM-DD_HHMMSS-<slug>.md, relative to the active workspace. Allows read-only commands and repo inspection. Infers the task to plan from the prompt, conversation context, or asks for clarification when needed. The output is the plan file and a brief reply.
- A developer wants a structured change plan before implementation, without risking accidental code edits.
- A project manager needs a clear, step-by-step plan file for a feature or refactor without making any changes.
- A user wants to convert a rough idea into an actionable plan for review or sharing.
- An agent needs a clear roadmap before proceeding, to reduce risk and scope creep.
What are this skill's strengths and limitations?
- Enforces a non-executive, planning-only workflow, preventing accidental changes.
- Produces structured, actionable plans with clear steps and risks.
- Works across domains, from code to any file-based task.
- Uses backend-aware relative paths for file writes, working across multiple backends.
- Saves plans to .hermes/plans/ instead of .dojo/plans/, conflicting with the description.
- No testing or validation steps are included.
- Uses an older timestamp format (YYYY-MM-DD_HHMMSS) that may not meet modern standards.
- No validation or linting of the generated plan is provided.
How do you install this skill?
This is an Agent Skill named 'plan', located at dojoagents/skills/built_in/plan/ in the repository. Installation involves placing the SKILL.md folder into your Agent Skills directory (e.g., ~/.claude/skills). The full collection installs via pip: uv pip install dojoagents, but specific installation of this skill is not separately documented. See the repository README for installation instructions.
How do you use this skill?
Trigger plan mode in a conversation (e.g., using the /plan command). The agent will then write a Markdown plan and save it under .hermes/plans/. You will receive a brief summary and the saved path. Example prompt: 'Plan the addition of a login feature, including steps and risks.' The agent will generate the plan file.