Prompt file imported from jcentner/chrome-dino (.github/prompts/implement.prompt.md). Fill in {{sliceNumber}} before use. Copyright stays with the author.

Implement

You are executing one slice of the approved implementation plan for chrome-dino. This prompt is a manual override entry point — the autonomous builder normally drives the slice loop.

Precondition

1. Read roadmap/state.md.
2. Verify Stage is executing. If not, stop and report the current stage.
3. Note the Active Slice number and the Implementation Plan path.

Input

Slice number (defaults to Active Slice from state.md): {{sliceNumber}}

Slice loop

This prompt mirrors the canonical slice loop in .github/agents/autonomous-builder.agent.md (section "The slice loop"). If the two ever disagree, the builder agent is authoritative.

Execute every step in order. Do not skip or reorder:

1. Read the slice from the implementation plan: files touched, scope, acceptance criteria, test strategy.
2. Invoke tester subagent with the spec. Tester writes tests without reading implementation (hook-enforced).
3. Implement the slice — edit only the files listed in the slice plan. If the slice grows, stop and update the plan first.
4. Run tests, then write the result explicitly with python3 .github/hooks/scripts/write-test-evidence.py pass|fail|n/a. On failure, apply the builder's error recovery: diagnose, fix (up to 2 attempts), otherwise record in docs/reference/tech-debt.md and move on.
5. Run post-implementation checks from the implementation plan (stale references, config traces, diagram/prose consistency).
6. Invoke reviewer subagent on the changed files — always, regardless of file count. Reviewer writes Review Verdict + findings to state.md.
7. Fix all Critical and Major findings before committing.
8. Commit: type(scope): description format, one slice per commit.
9. Advance state (single atomic update) in state.md:
   • Set Reviewer Invoked: yes, Review Verdict: <pass|needs-fixes|needs-rework>, Critical Findings: N, Major Findings: N, Committed: yes.
   • Increment Active Slice.
   • Reset slice evidence for the next slice: Tests Written: pending, Tests Pass: pending, Reviewer Invoked: pending, Review Verdict: pending, Critical Findings: 0, Major Findings: 0, Committed: pending.
   • On the next slice, rerun write-test-evidence.py as soon as you have a result. That helper stamps Evidence For Slice to the current Active Slice; session-gate blocks stop while Evidence For Slice != Active Slice — the hook-enforced backstop on the manual reset above.

The Stop hook (session-gate.py) blocks session end if Slice Evidence is incomplete during executing.

Autopilot behavior

Under autopilot, do not ask clarifying questions. Make evidence-based decisions, append rationale to ## Context in CURRENT-STATE.md, and record architectural choices as ADRs. Stage-gate and tool-guardrails hooks remain hard constraints.

Do not

• Edit files outside the slice's declared scope.
• Commit broken tests.
• Skip the reviewer. Every slice gets a reviewer pass, regardless of file count.
• Advance Active Slice before committing.