step3-execute-slice

Step 3: Execute Slices

Pipeline: Step 3 of 5 (Session 2 — Execution) When: After plan is hardened (Step 2), in a new agent session Model suggestion: Any model / Codex / Copilot Auto (10% token savings) — execution is mechanical; Codex is fast and focused Next Step: step4-completeness-sweep.prompt.md

Replace <YOUR-HARDENED-PLAN> with your hardened plan filename.

---

Read these files first:

1. docs/plans/AI-Plan-Hardening-Runbook.md (Execution Agent Prompt + Sections 10-11)
2. docs/plans/.md
3. .github/copilot-instructions.md

Now act as an EXECUTION AGENT (see the Execution Agent Prompt in the runbook).

<investigate_before_coding> Before writing code that depends on an existing file, read that file first. Never assume a method signature, type name, or import path — verify it by opening the file. If the plan references a file you haven't loaded, load it before coding against it. </investigate_before_coding>

<implementation_discipline> Only make changes specified in the current slice. Do not add features, refactor existing code, add abstractions, or create helpers beyond what the slice requires. Do not add error handling for scenarios that cannot occur within this slice's scope. Do not add docstrings, comments, or type annotations to code you did not change. The right amount of complexity is the minimum needed for the current slice. </implementation_discipline>

Execute the hardened plan one slice at a time, starting with Slice 1.

Before starting Slice 1, run a Pre-Execution Traceability Check:

• Check OpenBrain (if configured): search_thoughts("<plan topic>", project: "<YOUR PROJECT NAME>") — load prior gotchas, patterns, and lessons for the affected files/areas
• Check LiveGuard memories: Read .forge/liveguard-memories.jsonl if present — prior drift violations and incident history for this project
• Scan the spec's MUST acceptance criteria
• Verify each MUST criterion maps to at least one slice's validation gate
• If any MUST criterion has no corresponding validation gate, flag it and ask before proceeding

Before each slice, load its Context Files (including .github/instructions/*.instructions.md guardrails). When scaffolding new entities/services/tests, use the matching prompt template from .github/prompts/. Follow the validation loop exactly. Commit after each passed slice. If any gate fails or any ambiguity arises, pause and ask for clarification.

Re-anchor after each slice using the lightweight check (4 yes/no questions). Do a full re-anchor every 3rd slice or when a lightweight check flags a concern.

For [parallel-safe] slices:

• Note which Parallel Group they belong to
• After all slices in a group complete, run the Parallel Merge Checkpoint
• If any parallel slice fails, pause all slices in that group and report

After ALL slices pass, run the COMPLETENESS SWEEP (Section 6.1).

---