0%

Section 01: Tear Down Broken Pipeline

Status: Not Started Goal: Delete the broken Python analysis pipeline, scoring system, and schema. Preserve the reusable execution infrastructure (Steps 0-2 phase capture) in SKILL.md.

Success Criteria:

  • All Python files in .claude/skills/code-journey/ deleted except none (all go)
  • SCHEMA.md deleted (replaced by JSON schema in Section 02)
  • SKILL.md reduced to reusable skeleton (Steps 0-2 + modes preserved, scoring sections deleted)
  • rescore-all.sh deleted
  • All Python test files and golden data deleted
  • No dangling references to deleted files anywhere in the codebase
  • Satisfies mission criterion: “The ~5,000 lines of broken Python analysis pipeline are deleted”

Context: A dual-source TPR review (codex + gemini, 2026-04-09) independently confirmed that the Python analysis pipeline is fundamentally broken: flat RC summation hides path-dependent leaks, rc_state.py is dead code, effect_summaries.py has wrong COW effects, instruction metrics use a circular “ideal” definition, and the scoring system is a rubber stamp (18/20 at 10.0). The pipeline produces false confidence while real bugs go undetected. Deletion is the correct fix — the diagnostic tools in diagnostics/ are the real bug-finders.

Depends on: Nothing — this is Phase 0.

01.1 Delete Python Pipeline Files

File(s): .claude/skills/code-journey/

Delete ALL Python files, their tests, and supporting data:

  • Delete Python metric modules:

    • .claude/skills/code-journey/arc_metrics.py (452 lines — flat RC summation, broken)
    • .claude/skills/code-journey/rc_state.py (397 lines — dead code, never called)
    • .claude/skills/code-journey/instruction_metrics.py (192 lines — circular ideal metric)
    • .claude/skills/code-journey/control_flow_metrics.py (142 lines)
    • .claude/skills/code-journey/attribute_metrics.py (322 lines)
    • .claude/skills/code-journey/binary_metrics.py (128 lines)
    • .claude/skills/code-journey/effect_summaries.py (196 lines — wrong COW effects)
    • .claude/skills/code-journey/ir_parser.py (331 lines)
    • .claude/skills/code-journey/ir_parser_internal.py (252 lines)
    • .claude/skills/code-journey/ir_utils.py (106 lines)
    • .claude/skills/code-journey/extract_ir_from_results.py (104 lines)

  • Delete scoring and extraction scripts:

    • .claude/skills/code-journey/score.py (630 lines — arbitrary thresholds, rubber stamp)
    • .claude/skills/code-journey/extract-metrics.py (271 lines — wires broken arc_metrics)
    • .claude/skills/code-journey/rescore-all.sh (204 lines — batch re-scorer for broken system)

  • Delete the scoring schema:

    • .claude/skills/code-journey/SCHEMA.md (823 lines — score-centric, replaced by JSON schema in Section 02)

  • Delete Python test files and golden data:

    • .claude/skills/code-journey/tests/test_arc_metrics.py
    • .claude/skills/code-journey/tests/test_effect_summaries.py
    • .claude/skills/code-journey/tests/test_extract_metrics.py
    • .claude/skills/code-journey/tests/test_instruction_metrics.py
    • .claude/skills/code-journey/tests/test_rc_state.py
    • .claude/skills/code-journey/tests/test_attribute_metrics.py
    • .claude/skills/code-journey/tests/test_control_flow_metrics.py
    • .claude/skills/code-journey/tests/test_binary_metrics.py
    • .claude/skills/code-journey/tests/test_extract_ir.py
    • .claude/skills/code-journey/tests/golden/ (all golden test data)

  • Verify no dangling references:

    • Grep codebase for references to deleted filenames
    • Check .claude/commands/ for any references to deleted scripts
    • Check plans/code-journeys/overview.md for references to scoring tools

  • Subsection close-out (01.1) — MANDATORY before starting 01.2:

    • All tasks above are [x] and deletions verified
    • Update this subsection’s status in section frontmatter to complete
    • Run /improve-tooling retrospectively on THIS subsection
    • Run /sync-claude on THIS subsection — check whether code changes invalidated any CLAUDE.md, .claude/rules/*.md, or canon.md claims. If no API/command/phase changes, document briefly. Fix any drift NOW.
    • Repo hygiene check — run diagnostics/repo-hygiene.sh --check and clean any detected temp files.

01.2 Strip SKILL.md to Reusable Skeleton

File(s): .claude/skills/code-journey/SKILL.md

The current SKILL.md (692 lines) has reusable infrastructure (Steps 0-2: build run list, run both paths, capture phase dumps) and broken sections (scoring instructions, background agent scoring template, deep scrutiny framework). Strip it to a skeleton that Section 03 will rebuild.

  • Preserve SKILL.md frontmatter (name, description, argument-hint)

  • Preserve “CRITICAL: Scenario Preservation” section (NEVER modify existing .ori files)

  • Preserve “CRITICAL: Autonomous Execution” section (no user prompts)

  • Preserve “CRITICAL: Context Conservation” section (background agent delegation)

  • Preserve Step 0: Build the Run List (lines 66-108 — journey discovery, modes)

  • Preserve Step 1: Run Both Paths (lines 112-139 — eval + AOT execution)

  • Preserve Step 2: phase dump capture (lines 141-177 — env vars, temp files)

  • Preserve “Adding New Scenarios” section (lines 89-108 — gap-filling logic)

  • DELETE the scoring instructions section (lines 478-610)

  • DELETE the old background agent prompt template (tied to scoring)

  • DELETE references to SCHEMA.md, score.py, extract-metrics.py

  • REPLACE “CRITICAL: Schema Compliance” with a note: “Schema replaced — see Section 02 of plans/code-journey-rework/ for the JSON results schema”

  • Add a <!-- PLACEHOLDER: Section 03 will add the new orchestrator logic here --> marker where the new Steps 3-5 will go

  • Subsection close-out (01.2) — MANDATORY before marking section complete:

    • All tasks above are [x] and SKILL.md skeleton verified
    • Update this subsection’s status in section frontmatter to complete
    • Run /improve-tooling retrospectively on THIS subsection
    • Run /sync-claude on THIS subsection — check whether code changes invalidated any CLAUDE.md, .claude/rules/*.md, or canon.md claims. If no API/command/phase changes, document briefly. Fix any drift NOW.
    • Repo hygiene check — run diagnostics/repo-hygiene.sh --check and clean any detected temp files.

01.R Third Party Review Findings

  • None.

01.N Completion Checklist

  • All Python pipeline files deleted (~5,000 lines)
  • SCHEMA.md deleted
  • SKILL.md stripped to reusable skeleton
  • No dangling references to deleted files in codebase
  • timeout 150 ./test-all.sh green (Python files are not Rust-tested, but verify no build dependencies)
  • /tpr-review — dual-source review of teardown work
  • /impl-hygiene-review — verify no DRIFT (stale references), no WASTE (dead code left behind)
  • /improve-tooling section-close sweep — verify per-subsection retrospectives ran; add cross-subsection findings