Agent Skills: Planning instructions

When a planning mode (architect, planner-a, planner-b, planner-c) starts its work. This skill is *not* to be loaded for use with any other modes unless specifically called. Load with read_file on .kilocode/skills/planning-init/SKILL.md (ignore the absolute path in the location tag).

UncategorizedID: ScotterMonk/AgentAutoFlow/planning-init

Install this agent skill to your local

pnpm dlx add-skill https://github.com/ScotterMonk/AgentAutoFlow/tree/HEAD/.kilocode/skills/planning-init

Skill Files

Browse the full folder contents for planning-init.

Download Skill

Loading file tree…

.kilocode/skills/planning-init/SKILL.md

Skill Metadata

Name
planning-init
Description
When a planning mode (architect, planner-a, planner-b, planner-c) starts its work. This skill is *not* to be loaded for use with any other modes unless specifically called. Load with read_file on .kilocode/skills/planning-init/SKILL.md (ignore the absolute path in the location tag).

Planning instructions

Folders and files

  • Read skill-local guidance only when .kilocode/skills/planning-init/AGENTS.md is confirmed to exist; otherwise use root AGENTS.md for project-specific plan, backup, and completed-plan folders.
  • plans folder: Create if non-existent.
  • backups folder: Create if non-existent.
  • completed plans folder: Create if non-existent.
  • plans folder: plans/.
  • completed plans folder: plans/completed/.
  • backups folder: .kilocode/docs/old_versions/.
  • user query file: plans/p_[timestamp]_[short name]-user.md.
  • log file: plans/p_[timestamp]_[short name]-log.md.
  • plan file: plans/p_[timestamp]_[short name].md.

Variables

  • complexity.
  • testing type.

Workflow

Execute every step sequentially. Skip nothing. Steps:

1. Input

  1. Determine Plan status: Check for existence and content of log file and plan file to determine:
    • Plan = new: Continue past the next to possible status actions to "2) **Determine previous mode**".
    • Plan = incomplete:
      • Determine where progress was interrupted,
      • Use the new_task tool to switch to appropriate mode with message parameter containing only:
      • "Continue working on the plan in {plan_file}."
      • Critical to not include any other context.
    • Plan = finished: Move log file, plan file, user query file to completed plans folder, inform user, STOP.

1.5: Ambiguity gate

After capturing the user query, answer these silently:

  • Can I state the user's goal in one sentence without "probably" or "I assume"?
  • Is there exactly one plausible interpretation?
  • Are scope boundaries clear (what's in vs. out)?

If any answer is "no": Use resolve-ambiguity skill. Do not proceed to pre-planning until resolved.

  1. Determine previous mode: The following input instructions depend on which mode called the current mode (user or a planning mode):

    • IF Previous mode = query from user:
      • Action:
        • Capture input as user query.
        • Intuit the "why" in/for the user query.
        • Confirm your "why" guess with the user.
        • Modify user query to include the "why".
        • Save user query to user query file.
    • ELSE Previous mode = architect, planner-a or planner-b:
      • Action: Load into context:
        • plan & short plan name.
        • log file name.
        • user query & user query file name.
        • complexity.
      • Validation: If context is incomplete, alert user and halt immediately.
      • **Skip remaining instructions in this ### 1. Input section down to continue at ### 2. Pre-planning.
  2. Short plan name: If empty short plan name, derive from user query.

    • Create/use user query file.
    • Create/use log file.
    • Create/use plan file.
  3. Configuration: If following 3 config items are empty: Recommend best complexity for this plan: One Phase (Tiny/Small), One Phase (Small/Med), Few Phases (Med), Multi-Phase (Large). Set testing type as Use what is appropriate per task

2. Pre-planning

  1. Search: Search for similar planning documents and architectural decisions.
  2. Recall: Retrieve project history/memory.
  3. Risk: Identify potential challenges.
  4. Analysis: Define problem, intent, scope, constraints, and dependencies.

3. Analysis 2:

  • Double-check problem (user query file), intent, scope, constraints, and dependencies.
  • Find and inform user of redundancies or other issues before continuing.