Prism Loop

PRISM Workflow Loop

TDD-driven workflow orchestration using the Ralph Wiggum self-referential loop pattern.

Quick Start

1. Run *prism-loop [your context/prompt]
2. SM agent reviews previous notes and drafts story
3. QA agent writes failing tests (TDD RED)
4. Red gate pauses for /prism-approve
5. DEV agent implements tasks (TDD GREEN)
6. QA verifies green state, green gate completes

When to Use

• User wants to run the PRISM core development cycle
• Starting a new story implementation with TDD
• Need automated workflow progression with gates

How It Works

1. Stop Hook intercepts session exit and re-injects the next step instruction
2. Agent steps auto-progress (SM → QA → DEV)
3. Gate steps pause for /prism-approve (or /prism-reject at red_gate)
4. Validation runs tests to verify TDD state (RED = fail, GREEN = pass)

Workflow Steps (7 steps)

Commands

*prism-loop [prompt]

Start the PRISM workflow loop.

python "${CLAUDE_PLUGIN_ROOT}/skills/prism-loop/scripts/setup_prism_loop.py" --session-id "${CLAUDE_SESSION_ID}" "$ARGUMENTS"

The prompt provides context to the SM agent for planning.

Example:

*prism-loop implement user authentication feature

*prism-approve

Approve the current gate and advance to next phase.

python "${CLAUDE_PLUGIN_ROOT}/skills/prism-loop/scripts/prism_approve.py"

• At red_gate: Proceeds to GREEN phase (implementation)
• At green_gate: Completes workflow

*prism-reject

Reject at red_gate and loop back to planning (step 1).

python "${CLAUDE_PLUGIN_ROOT}/skills/prism-loop/scripts/prism_reject.py"

Only valid at red_gate. Use when tests need redesign.

*prism-status

Check current workflow state.

python "${CLAUDE_PLUGIN_ROOT}/skills/prism-loop/scripts/prism_status.py"

Shows progress through all 7 steps.

*cancel-prism

Cancel the active workflow.

python "${CLAUDE_PLUGIN_ROOT}/skills/prism-loop/scripts/cancel_prism_loop.py"

Removes state file and stops the loop.

Requirements Traceability

The workflow enforces a traceability chain to prevent silent requirement drops:

User Request → Requirements → Acceptance Criteria → Tests
              (REQ-1, REQ-2)   (AC-1 → REQ-1)      (test_ac1 → AC-1)

Capture and Validation Points

Test Mapping Conventions

For the trace audit to find test coverage:

If an AC has no mapped test, the workflow blocks with "SILENT DROP DETECTED".

Trace Matrix at Red Gate

The red_gate displays a trace matrix for human verification:

## Trace Matrix
| REQ | AC | Test | Status |
|-----|-----|------|--------|
| REQ-1 | AC-1 | test_ac1_login | COVERED |
| REQ-2 | AC-2 | test_ac2_logout | COVERED |

This ensures no requirement silently disappears during implementation.

TDD Validation

The stop hook validates before advancing:

• write_failing_tests → Tests must FAIL (assertion errors, not syntax errors) + trace audit passes
• implement_tasks → All tests must PASS
• verify_green_state → Tests + lint must pass

Claude cannot "think" it's done - the hook runs tests to verify. The trace audit ensures every acceptance criterion has test coverage before proceeding to implementation.

State File

Located at .claude/prism-loop.local.md

Tracks:

• current_step: Active step
• current_step_index: Position (0-6)
• story_file: Path to story file (set after draft_story)
• paused_for_manual: True at gates

Integration

The stop hook is registered in hooks/hooks.json:

{
  "Stop": [{
    "matcher": "*",
    "hooks": [{
      "type": "command",
      "command": "python ${CLAUDE_PLUGIN_ROOT}/hooks/prism_stop_hook.py"
    }]
  }]
}

Example Session

# Start workflow
*prism-loop implement login feature

# SM agent runs planning phases automatically
# QA writes failing tests
# Stop hook blocks until tests fail correctly

# At red_gate - approve to continue
*prism-approve

# DEV implements until tests pass
# QA verifies

# At green_gate - complete
*prism-approve

# Done!

Triggers

This skill activates when you mention:

• "prism loop" or "prism workflow"
• "start development cycle"
• "TDD workflow" or "core development cycle"
• "/prism" or "/prism-loop"

Version: 3.4.0 Last Updated: 2025-02-06