Execute Plan

Relies on BDK foundation (STARTUP_INSTRUCTIONS.md) for project context and MCP tool preference.

!python3 ${CLAUDE_PLUGIN_ROOT}/scripts/inject.py --chain ${CLAUDE_PLUGIN_ROOT}/fragments/tool-tiers/impact.chain.json

For each <!-- INJECT: <name> --> marker below, run python3 ${CLAUDE_PLUGIN_ROOT}/scripts/inject-rules.py <name> and substitute its stdout in place of the marker before acting on it.

Step 0: Create Task List

Before executing, create TaskList from plan:

• One TaskCreate per implementation task
• Set status in_progress on start, completed when done

Step 1: Execute the Plan

$ARGUMENTS

Per task:

1. Has **Test cases:** section?

   • YES → invoke /bdk:test-driven-development before any implementation. Pass test bullets + implementation desc. No implementation until GATE 2 (RED verified).
   • NO → implement directly (e.g. refactor-only)

2. After completed, check context:

   • < 40% → next task
   • ≥ 40% → pause, write:

     ⚠️ Context at XX% — consider running /bdk:save-progress [slug] before continuing. Ask: "Continue or save progress first?" Wait for response.

3. Blocked or unclear → stop, ask user

Step 2: Verify and Fix

All tasks done → inject project tools context:

• Test tools: !python3 ${CLAUDE_PLUGIN_ROOT}/scripts/get_settings.py test-tools
• Lint tools: !python3 ${CLAUDE_PLUGIN_ROOT}/scripts/get_settings.py lint-tools

Run verification agents in parallel:

2a. Run Tests

Delegate to test-runner agent using injected test command above. Tests fail → fix, re-run. Repeat until pass.

2b. Run Static Analysis

Delegate to static-analyse agent using injected lint command above. Issues found → fix, re-run. Repeat until clean.

Rules

• Main agent fixes issues, subagents only run
• Max 3 fix-and-rerun cycles per step before asking user
• NEVER hardcode test or lint commands — use injected values, fall back to detecting from project context