sase-plan

name	sase_plan
description	Create an implementation plan. Use instead of plan mode (which is disabled).

Use this skill when you need to plan before implementing. This replaces Claude's native plan mode, which is disabled.

Instructions

1. Explore and understand the problem thoroughly.

2. Choose the plan tier before writing:

   • Use tale for work that one follow-up coding agent can implement as a single plan.
   • Use epic when the work should be split into phases that distinct agents can complete. Declare every phase dependency explicitly (so we can support parallel work if needed/desirable). Every phase in an epic plan file MUST have a unique slug ID.

3. Write a self-contained plan to sase_plan_<name>.md (descriptive underscore name).

   • The file must start at byte 0 with valid YAML frontmatter that contains a single tier: <tier> property, where <tier> is either tale or epic.

4. Validate (with --explain), edit, and revalidate (without --explain):

   The first validation run with --explain prints the expected schema and all diagnostics. Use that information to edit the plan file. Then rerun validation without --explain to check that the file is now valid. Continue until validation exits successfully. Do not propose a plan that has not passed validation.

   sase plan validate sase_plan_<name>.md --explain
   # ... edit the file to fix all reported issues ...
   sase plan validate sase_plan_<name>.md
   # ... repeat until validation exits successfully ...
   

5. Submit the validated plan:

   sase plan propose sase_plan_<name>.md
   

   Submission consumes the scratch plan into SASE's durable plan archive, writes a handoff marker, and sends SIGTERM to the current agent runner process group. The runner treats that signal as an intentional handoff: it creates the tier-specific PlanApproval or EpicApproval gate, waits mechanically for a terminal response, and continues with feedback or the approved follow-up. Do not expect the current provider turn to return normally after a successful proposal, and do not poll response files yourself.