Prompt file imported from Crispy13/VarDict_RS (.github/prompts/work_for_goal.prompt.md). Copyright stays with the author.

Workflow

You MUST perform the following workflow:

Definitions

• Mission goal: the higher-level objective described in copilot-office/<mission-name>/copilot-project-plan.md.
• Active plan: copilot-office/<mission-name>/copilot-active-plan.md, which tracks the currently active goal, the working state needed to continue it, and the high-level slices or expected stages for that goal.
• Stage plan: copilot-office/<mission-name>/copilot-stage-plan.md, which defines the next actionable stage to execute.
• Active goal: the concrete goal you are currently trying to achieve in this run. It may be a user-provided subgoal or, if no subgoal is active, the mission goal itself.
• Stage objective: the specific execution slice of the active goal described in the current stage plan.

0. Resolve The Active Goal

• First resolve copilot-office/<mission-name>/ using ../instructions/project_context.instructions.md.
• Then determine the active goal using this priority:

1. If the user gives an explicit request in the current prompt, that request is the active goal.
2. Otherwise, if session history or copilot-office/<mission-name>/copilot-active-plan.md already shows an unresolved active goal, continue that active goal.
3. Otherwise, use the mission goal from copilot-office/<mission-name>/copilot-project-plan.md as the active goal.

• Do not confuse the mission goal with the active goal. The mission goal is the larger objective; the active goal is what you are currently pursuing.
• If the active goal is vague or cannot be determined safely, ask for clarification using vscode_askQuestions instead of making assumptions.
• If copilot-office/<mission-name>/copilot-active-plan.md is missing, stale, or no longer matches the active goal, run a custom agent named exactly 'Plan' as subagent to produce updated active-plan content, then write that content to copilot-office/<mission-name>/copilot-active-plan.md in the main agent.
• In that active-plan generation step, break the active goal into several high-level slices or expected stages when useful.
• Do NOT refresh the active plan on every loop by default. Refresh it only when the active goal, stage slicing, blocker model, key decisions, or overall strategy changed materially, or when the active plan is missing, stale, or inconsistent.
• Do not move to step 1 until the active goal is resolved safely and the active plan is usable, or a genuine blocker is reported.

1. Plan The Next Stage

1. Run a custom agent named exactly 'Plan' as subagent to update or generate the next stage plan:

• If the Plan subagent is unavailable, stop and report that blocker instead of silently substituting another agent.

1. Confirm the resolved mission folder and ignore unrelated mission folders unless the user explicitly asks for cross-mission work.

2. Understand the mission context referring to ../instructions/project_context.instructions.md.

3. Use the slices or expected stages already recorded in copilot-office/<mission-name>/copilot-active-plan.md when they are still valid. Re-slice only if the active plan is missing that structure or it is no longer valid.

4. Produce the next stage-plan content for copilot-office/<mission-name>/copilot-stage-plan.md and return it to the main agent.

5. Update copilot-office/<mission-name>/copilot-stage-plan.md in the main agent with the stage-plan content returned by the Plan subagent.

2. Execute The Current Stage

Run a subagent to execute the stage plan:

• When delegating to subagents, include these:
  1. the resolved mission name
  2. the stage plan path: copilot-office/<mission-name>/copilot-stage-plan.md (direct order for subagent)
  3. project context instruction path: ../instructions/project_context.instructions.md
• Tell them to stay within that mission folder and its related shared docs.
• After completing the current stage, reassess the active goal.
• Update copilot-office/<mission-name>/copilot-active-plan.md only when it truly needs modification.
• Run the custom agent named exactly Plan as subagent to produce updated active-plan content only when the active goal changed, the expected stage slices changed, the blocker model changed materially, key decisions changed, or the current strategy is no longer valid.
• If the active goal is not yet resolved and no genuine blocker exists, return to step 1 with updated state and plan the next stage. Before returning to step 1 for another loop, read ./work_for_goal.prompt.md again to refresh this workflow and reduce drift across repeated iterations.
• Keep iterating stages until the active goal is resolved or blocked.

3. Reporting

Once you reach a natural stopping point, Run a subagent to provide a comprehensive report detailing the resolved mission folder, whether it came from current session history or the inspection step, the mission goal, the active goal, completed stages, remaining backlog, any plan cleanup performed, and any uncovered issues or blockers.