TEMPLATE

<!--
This is the template every prompt for an Executor session fills out. Prompts
fall into two categories:

- **execution/** — task execution prompts (one per TSK-NNN). Generated by the
  Planner during a planning session, saved at `.ai/prompts/execution/<phase>/TSK-NNN.md`,
  and run in a separate Executor session.
- **planning/** — replanning or planning-event prompts. Generated when the
  roadmap structure changes (renumbering, adding/removing tasks, convention
  adoption). Saved at `.ai/prompts/planning/YYYY-MM-DD-<slug>.md`.

All six sections below are mandatory regardless of category. Do not delete or rename them.
See `.ai/workflows/TASK-PLANNER.md` for the task planning workflow.
-->

<context>
<!-- Which repo, which phase, which task, which files the Executor needs to
     load into context. Keep to 3-5 lines.

     Example:
     Repo: AI-first backend project. Docs at @CLAUDE.md, @docs/engineering/STRUCTURE.md.
     Phase N of the build plan, TSK-NNN — <short title>.
     Builds on TSK-MMM (already merged). -->
</context>

<task>
<!-- One or two sentences. Imperative, active voice.
     Example: "Create X and Y at the repo root." -->
</task>

<requirements>
<!-- Specific, verifiable items the result must have.
     Group in sub-blocks if covering multiple artifacts.
     Use concrete names, paths, and values — not abstractions.

     Example:
     **`.gitignore`:**
     - Covers: virtual environments, build artifacts, IDE files, secret files.
     - Grouped by category.
     - The dependency lockfile is NOT listed (it is tracked by git).

     **`README.md` (minimal):**
     - Project name and one-line description.
     - Install command, test command.
     - Link to `@docs/INDEX.md`.
-->
</requirements>

<constraints>
<!-- Limits and prohibitions. Keep short — only when the Executor's natural
     tendency would violate them.

     Example:
     - Do not modify files outside the scope above.
     - Do not commit — the human handles commits.
     - Do not run package installs unless explicitly requested. -->
</constraints>

<deliverable>
<!-- Optional. Only include if the output needs a specific format.

     Example: "Print a short summary when done: which files were created
     and any decision you made that differs from the requirements above." -->
</deliverable>

<runtime-decisions>
<!-- Decisions made during prompt execution when the prompt didn't specify a
     detail. The Executor appends entries here (not the user). Keep prompts
     self-contained for future reuse by preserving these decisions.

     Format for each entry:

     ### <short topic name>
     **Asked:** <the question the Executor asked the user>
     **Answered:** <concise summary of the user's response>
     **Applied as:** <how it was translated into the solution>
-->
</runtime-decisions>