codex-genesis-harness 0.1.0 → 0.1.4

package/.codex/skills/{project-genesis-harness → genesis-harness}/scripts/list-changed-files.sh

@@ -1,12 +1,12 @@
-#!/usr/bin/env bash
-set -euo pipefail
-
-root="${1:-.}"
-cd "$root"
-
-if git rev-parse --is-inside-work-tree >/dev/null 2>&1; then
-  git status --short
-else
-  echo "Not a git repository; changed files cannot be listed with git."
-fi
-
+#!/usr/bin/env bash
+set -euo pipefail
+
+root="${1:-.}"
+cd "$root"
+
+if git rev-parse --is-inside-work-tree >/dev/null 2>&1; then
+  git status --short
+else
+  echo "Not a git repository; changed files cannot be listed with git."
+fi
+

package/.codex/skills/genesis-harness/scripts/offload-log.sh

@@ -0,0 +1,72 @@
+#!/usr/bin/env bash
+# ==============================================================================
+# Genesis Codex Harness — Tool Call Offloading Script
+# ==============================================================================
+# Captures standard output & error of any command, saves it to a durable log file,
+# and prints a trimmed head-and-tail summary if it exceeds a line count threshold.
+# Protects the AI context window against Context Rot.
+# ==============================================================================
+
+set -uo pipefail
+
+THRESHOLD_LINES=50
+SCRATCH_DIR="scratch/tool_outputs"
+
+usage() {
+  echo "Usage: $0 [--threshold lines] <command> [args...]" >&2
+  exit 2
+}
+
+if [ "$#" -eq 0 ]; then
+  usage
+fi
+
+if [ "$1" = "--threshold" ]; then
+  THRESHOLD_LINES="$2"
+  shift 2
+fi
+
+if [ "$#" -eq 0 ]; then
+  usage
+fi
+
+mkdir -p "$SCRATCH_DIR"
+
+# Generate unique log file path
+TIMESTAMP="$(date +%Y%m%d_%H%M%S)"
+RAND_ID=$((RANDOM % 10000))
+LOG_FILE="$SCRATCH_DIR/output_${TIMESTAMP}_${RAND_ID}.log"
+
+# Execute command and capture output
+set +e
+"$@" > "$LOG_FILE" 2>&1
+EXIT_CODE=$?
+set -e
+
+# Calculate output size
+LINE_COUNT=$(wc -l < "$LOG_FILE")
+
+if [ "$LINE_COUNT" -le "$THRESHOLD_LINES" ]; then
+  # Print normally if within threshold
+  cat "$LOG_FILE"
+else
+  # Offload and print head/tail summary
+  HEAD_LINES=$((THRESHOLD_LINES / 2))
+  TAIL_LINES=$((THRESHOLD_LINES / 2))
+
+  echo "=============================================================================="
+  echo "==> [OFFLOADED] Command output exceeded threshold ($LINE_COUNT lines > $THRESHOLD_LINES lines)."
+  echo "==> Full log written to disk: file://$(pwd)/$LOG_FILE"
+  echo "=============================================================================="
+  
+  head -n "$HEAD_LINES" "$LOG_FILE"
+  
+  echo ""
+  echo "... [TRUNCATED $(($LINE_COUNT - THRESHOLD_LINES)) LINES] ..."
+  echo "Read file://$LOG_FILE to view the full execution history."
+  echo ""
+  
+  tail -n "$TAIL_LINES" "$LOG_FILE"
+fi
+
+exit $EXIT_CODE

package/.codex/skills/{project-genesis-harness → genesis-harness}/scripts/run-verification.sh

@@ -1,47 +1,47 @@
-#!/usr/bin/env bash
-set -euo pipefail
-
-root="${1:-.}"
-cd "$root"
-
-run_if() {
-  local label="$1"
-  shift
-  echo "==> $label"
-  "$@"
-}
-
-if [ -f package.json ]; then
-  if command -v npm >/dev/null 2>&1; then
-    npm run lint --if-present
-    npm run typecheck --if-present
-    npm test --if-present
-    npm run build --if-present
-  fi
-elif [ -f pyproject.toml ] || [ -f requirements.txt ]; then
-  command -v pytest >/dev/null 2>&1 && run_if "pytest" pytest -q
-  command -v ruff >/dev/null 2>&1 && run_if "ruff" ruff check .
-  command -v mypy >/dev/null 2>&1 && run_if "mypy" mypy .
-elif [ -f Cargo.toml ]; then
-  run_if "cargo test" cargo test
-  run_if "cargo clippy" cargo clippy -- -D warnings
-elif [ -f go.mod ]; then
-  run_if "go test" go test ./...
-elif [ -f composer.json ]; then
-  command -v composer >/dev/null 2>&1 && composer test || true
-else
-  echo "No known verification command detected. Add project-specific commands to .planning/SMOKE_TESTS.md."
-fi
-
-for check in \
-  check-required-planning-files.sh \
-  check-task-tracking.sh \
-  check-no-debug-logs.sh \
-  check-spec-changelog.sh \
-  check-architecture-boundaries.sh
-do
-  script_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
-  if [ -x "$script_dir/$check" ]; then
-    run_if "$check" "$script_dir/$check" "$root"
-  fi
-done
+#!/usr/bin/env bash
+set -euo pipefail
+
+root="${1:-.}"
+cd "$root"
+
+run_if() {
+  local label="$1"
+  shift
+  echo "==> $label"
+  "$@"
+}
+
+if [ -f package.json ]; then
+  if command -v npm >/dev/null 2>&1; then
+    npm run lint --if-present
+    npm run typecheck --if-present
+    npm test --if-present
+    npm run build --if-present
+  fi
+elif [ -f pyproject.toml ] || [ -f requirements.txt ]; then
+  command -v pytest >/dev/null 2>&1 && run_if "pytest" pytest -q
+  command -v ruff >/dev/null 2>&1 && run_if "ruff" ruff check .
+  command -v mypy >/dev/null 2>&1 && run_if "mypy" mypy .
+elif [ -f Cargo.toml ]; then
+  run_if "cargo test" cargo test
+  run_if "cargo clippy" cargo clippy -- -D warnings
+elif [ -f go.mod ]; then
+  run_if "go test" go test ./...
+elif [ -f composer.json ]; then
+  command -v composer >/dev/null 2>&1 && composer test || true
+else
+  echo "No known verification command detected. Add project-specific commands to .planning/SMOKE_TESTS.md."
+fi
+
+for check in \
+  check-required-planning-files.sh \
+  check-task-tracking.sh \
+  check-no-debug-logs.sh \
+  check-spec-changelog.sh \
+  check-architecture-boundaries.sh
+do
+  script_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+  if [ -x "$script_dir/$check" ]; then
+    run_if "$check" "$script_dir/$check" "$root"
+  fi
+done

package/.codex/skills/genesis-harness/scripts/run-verify-loop.sh

@@ -0,0 +1,75 @@
+#!/usr/bin/env bash
+# ==============================================================================
+# Genesis Codex Harness — Ralph Loop (Verify-Fix) Orchestrator
+# ==============================================================================
+# Executes verification commands, tracks iteration counts to prevent infinite loops,
+# and logs diagnostic failure reports to disk for step resumption.
+# Enforces the self-verification safety invariant.
+# ==============================================================================
+
+set -uo pipefail
+
+MAX_ITERATIONS=5
+PLANNING_DIR=".planning"
+LOOP_COUNT_FILE="$PLANNING_DIR/VERIFY_LOOP_COUNT"
+FAILURE_LOG="$PLANNING_DIR/LAST_VERIFICATION_FAILURE.md"
+
+usage() {
+  echo "Usage: $0 [--max-iterations N] <verification_command> [args...]" >&2
+  exit 2
+}
+
+if [ "$#" -eq 0 ]; then
+  usage
+fi
+
+if [ "$1" = "--max-iterations" ]; then
+  MAX_ITERATIONS="$2"
+  shift 2
+fi
+
+if [ "$#" -eq 0 ]; then
+  usage
+fi
+
+mkdir -p "$PLANNING_DIR"
+
+# Initialize loop count
+CURRENT_ITERATION=1
+if [ -f "$LOOP_COUNT_FILE" ]; then
+  CURRENT_ITERATION=$(cat "$LOOP_COUNT_FILE")
+  CURRENT_ITERATION=$((CURRENT_ITERATION + 1))
+fi
+
+echo "$CURRENT_ITERATION" > "$LOOP_COUNT_FILE"
+
+echo "==> [VERIFY LOOP] Iteration $CURRENT_ITERATION of $MAX_ITERATIONS"
+
+if [ "$CURRENT_ITERATION" -gt "$MAX_ITERATIONS" ]; then
+  echo "=============================================================================="
+  echo "==> [BLOCKED] Verify-Fix loop exceeded maximum iteration count ($MAX_ITERATIONS)."
+  echo "==> Potentially stuck in an infinite loop. Halting for human intervention."
+  echo "=============================================================================="
+  rm -f "$LOOP_COUNT_FILE"
+  exit 99
+fi
+
+# Execute verification command
+set +e
+"$@" > "$FAILURE_LOG" 2>&1
+EXIT_CODE=$?
+set -e
+
+if [ $EXIT_CODE -eq 0 ]; then
+  echo "==> [VERIFY LOOP] Pass! Verification completed successfully."
+  rm -f "$LOOP_COUNT_FILE"
+  rm -f "$FAILURE_LOG"
+else
+  echo "=============================================================================="
+  echo "==> [FAILURE] Verification failed with exit code $EXIT_CODE."
+  echo "==> Failure diagnostic written to: file://$(pwd)/$FAILURE_LOG"
+  echo "==> Please inspect the log, fix the root cause, and re-run verification."
+  echo "=============================================================================="
+fi
+
+exit $EXIT_CODE

package/.codex/skills/{project-genesis-harness → genesis-harness}/scripts/update-state.sh

@@ -1,33 +1,33 @@
-#!/usr/bin/env bash
-set -euo pipefail
-
-usage() {
-  echo "Usage: $0 <root> <current-phase> <active-work> <last-completed-task> <next-task> <latest-verification>" >&2
-  exit 2
-}
-
-root="${1:-}"
-phase="${2:-}"
-active="${3:-}"
-last="${4:-}"
-next="${5:-}"
-verification="${6:-}"
-
-[ -n "$root" ] && [ -n "$phase" ] && [ -n "$active" ] && [ -n "$last" ] && [ -n "$next" ] && [ -n "$verification" ] || usage
-
-cd "$root"
-mkdir -p .planning
-
-cat > .planning/STATE.md <<EOF
-# State
-
-Current project state: [~] Active
-Current phase: $phase
-Current feature or bug: $active
-Last completed task: $last
-Next task: $next
-Blocked items: None recorded
-Latest verification result: $verification
-EOF
-
-echo ".planning/STATE.md"
+#!/usr/bin/env bash
+set -euo pipefail
+
+usage() {
+  echo "Usage: $0 <root> <current-phase> <active-work> <last-completed-task> <next-task> <latest-verification>" >&2
+  exit 2
+}
+
+root="${1:-}"
+phase="${2:-}"
+active="${3:-}"
+last="${4:-}"
+next="${5:-}"
+verification="${6:-}"
+
+[ -n "$root" ] && [ -n "$phase" ] && [ -n "$active" ] && [ -n "$last" ] && [ -n "$next" ] && [ -n "$verification" ] || usage
+
+cd "$root"
+mkdir -p .planning
+
+cat > .planning/STATE.md <<EOF
+# State
+
+Current project state: [~] Active
+Current phase: $phase
+Current feature or bug: $active
+Last completed task: $last
+Next task: $next
+Blocked items: None recorded
+Latest verification result: $verification
+EOF
+
+echo ".planning/STATE.md"

package/.codex/skills/genesis-harness/templates/harness-workflow-template.md

@@ -0,0 +1,10 @@
+# Harness Workflow Template
+
+## Intent
+## Failing Test
+## Fixture
+## Contract
+## Implementation
+## Verification
+## Memory Update
+

package/.codex/skills/genesis-harness-engineering/SKILL.md

@@ -0,0 +1,159 @@
+---
+name: harness-engineering-skill
+description: "Evolve the Codex harness itself: verification loops, repository memory, test-first scaffolds, resumability, observability, and autonomous workflow reliability. Use for changes to this repository's skill system or harness architecture."
+---
+
+# Harness Engineering Skill
+
+**Automated evolution, verification scaling, and structural safety loop engineering for Codex developer environments**
+
+---
+
+## Purpose
+
+To provide a robust, self-healing, and deterministic development environment for autonomous coding agents (Codex). Harness Engineering shifts the focus from writing soft, non-deterministic natural language prompts to building hard, mechanically-enforced environmental constraints (feedback loops, test runners, custom lints, and state managers) that prevent agentic drift and guarantee software quality.
+
+By scaling and optimizing this harness, we achieve two primary objectives:
+1.  **Extreme Feedback Velocity**: Compressing the time from code generation to execution feedback (TDD loop) to allow rapid self-correction.
+2.  **Safety & Invariance Enforcement**: Ensuring that all changes satisfy strict codebase constraints (lints, typechecks, API contracts, and E2E journeys) before completion is declared.
+
+---
+
+## When to use
+
+Use this skill when:
+-   **Modifying Core Verification Logic**: Adding new checks to `scripts/verify.sh` or expanding `scripts/run-evals.sh`.
+-   **Evolving Codex Skills**: Modifying existing `.codex/skills/` metadata, workflows, templates, or checklists.
+-   **Updating API, Agent, or Event Contracts**: Defining new schemas, request/response fixtures, or event topologies.
+-   **Extending Observability**: Adding tracking files, decision-logging formats, or failure telemetry inside `observability/`.
+-   **Configuring Distribution Channels**: Modifying the npm CLI wrapper (`bin/genesis-harness.js`), installers (`install.sh`), or CI/CD pipelines.
+
+---
+
+## When NOT to use
+
+Do not use this skill for:
+-   Implementing application-specific business logic or customer features.
+-   General codebase research that does not affect the agentic operating environment.
+-   One-off scripting tasks unrelated to the reliability and verification loop of Codex.
+
+---
+
+## Inputs required
+
+To evolve the harness, Codex must have access to:
+1.  **Current Operating Memory**: `.codebase/CURRENT_STATE.md` and `.codebase/MODULE_INDEX.md`.
+2.  **Harness Test Matrix**: `.codebase/TEST_MATRIX.md` showing existing verification paths.
+3.  **Target Workflow Specs**: The contract or script being evolved (e.g., `bin/genesis-harness.js`, `scripts/verify.sh`).
+4.  **Verification Gap**: The failure scenario or regression that is not currently caught by the test suite.
+
+---
+
+## Outputs required
+
+Every harness evolution must deliver:
+1.  **Failing Regression Test**: A new verification case added to `scripts/verify.sh` or `scripts/run-evals.sh` that fails *before* implementation.
+2.  **Atomic Implementation Change**: The minimum code, script, or structural edit required to pass the test.
+3.  **Clean Run Log**: Execution proof showing `scripts/verify.sh` and `scripts/run-evals.sh` passing cleanly with exit code `0`.
+4.  **Normalized Script Files**: LF line ending assurance on all modified `.sh` files to prevent Windows compatibility bugs.
+5.  **Durable Memory Update**: A recovery checkpoint written to `.codebase/RECOVERY_POINTS.md`.
+
+---
+
+## Required tests
+
+-   **Red-Green-Refactor Flow**: A failing assertion must be added to the test suite (`scripts/verify.sh` or `scripts/run-evals.sh`) prior to code modification.
+-   **Command Integrity Check**: The CLI `genesis-harness` must be verified using `npm run pack:check` and local execution tests in a separate sandbox.
+-   **Cross-Platform Validation**: All scripts must be checked for line-ending and interpreter syntax compatibility across POSIX/Bash environments on Windows/WSL and Linux.
+
+---
+
+## Required fixtures
+
+-   **Contract Templates**: Up-to-date `.json` schema templates for all agent, event, API, and UI interaction contracts.
+-   **Scaffold Fixtures**: Reusable sample files representing typical project states (like the `sample-feature` and `sample-bug` structures generated in test sandboxes).
+-   **Expected Validation Outputs**: JSON or Markdown files representing the expected logs or decision outputs of a successful run.
+
+---
+
+## Required contract updates
+
+-   **CLI-Skill Protocol**: Any change to CLI arguments, environment variable mappings, or directory output paths must be immediately reflected in `.codex-plugin/plugin.json`.
+-   **Agent 캐릭터 Contracts**: Changes to agent execution behavior must be documented in `contracts/agents/` schemas.
+-   **Event Topologies**: Updates to the event broker simulation must be verified against `contracts/events/` schemas.
+
+---
+
+## Required codebase map updates
+
+-   **Module Index Sync**: Newly added skills or templates must be registered in `.codebase/MODULE_INDEX.md` and `.codex/SKILLS_INDEX.md`.
+-   **Architecture Log**: Major design shifts in the harness workflow must be logged as Architecture Decision Records under `.codebase/decisions/` or `.planning/decisions/`.
+-   **State Transition**: Recalculate and update the architecture and quality scores inside `.codebase/CURRENT_STATE.md`.
+
+---
+
+## Token saving rules
+
+-   **Strict Ephemerality**: Treat the model's context window as volatile. Write key state updates to on-disk files (`task.md` and `.codebase/` memory) between steps instead of carrying huge logs in the prompt history.
+-   **On-Demand Loading**: Never load multiple `SKILL.md` files at once. Read the `.codebase/context/` summaries to target specific sub-directories.
+-   **Keep Commands Silent**: Suppress verbose compiler or script outputs (`>/dev/null`) unless a failure occurs, saving significant context tokens.
+
+---
+
+## Acceptance criteria
+
+A harness modification is accepted *only* when:
+-   [x] The new check fails first in a fresh, isolated run of the verification pipeline.
+-   [x] The minimal change is implemented, resulting in a successful pass of `verify.sh` and `run-evals.sh` with exit code `0`.
+-   [x] All changed `.sh` files are validated to have LF line endings and parse cleanly without syntax warnings.
+-   [x] CLI dry-run package checks (`npm run pack:check`) confirm package integrity.
+-   [x] The evolution is documented in `.codebase/RECOVERY_POINTS.md` with exact recovery instructions.
+
+---
+
+## Common mistakes
+
+-   **Writing prose instructions**: Attempting to guide the model using soft natural-language descriptions in READMEs instead of writing hard validation scripts.
+-   **Duplicating context maps**: Storing identical configuration maps across multiple skills instead of importing standard `.codebase/` definitions.
+-   **CRLF Line Ending Traps**: Checking in scripts with Windows line-endings (`\r\n`), causing execution syntax errors in WSL/POSIX shells.
+-   **Generator-Evaluator Co-location**: Allowing the agent to grade its own output format using regex within its prompt instead of executing an external compiler or validator command.
+
+---
+
+## Recovery workflow
+
+If a harness evolution breaks the local environment or causes test failures:
+1.  **Isolate the Failing State**: Stop all active background processes and subagents.
+2.  **Revert to Safety**: Run `git checkout -- <file>` to revert to the last committed stable state.
+3.  **Read Recovery Point**: Inspect `.codebase/RECOVERY_POINTS.md` to identify the nearest stable checkpoint.
+4.  **Write Minimal Reproducer**: Create a single, isolated script in the `scratch/` directory that demonstrates only the failure.
+5.  **Re-attempt TDD**: Implement the fix on top of the reproducer before integrating back into the main harness.
+
+---
+
+## Core Invariants: The PEV Loop
+
+Every agentic activity within this harness must strictly adhere to the **Plan-Execute-Verify (PEV)** pipeline:
+
+```txt
+   +---------------------------------------------------+
+   |             Plan Phase (DoR Verified)             |
+   |  - Research patterns and write on-disk PLAN.md    |
+   +---------------------------------------------------+
+                             |
+                             v
+   +---------------------------------------------------+
+   |            Execute Phase (TDD Enforced)           |
+   |  - Create failing unit test/verification check    |
+   |  - Write minimal code to pass verification       |
+   +---------------------------------------------------+
+                             |
+                             v
+   +---------------------------------------------------+
+   |             Verify Phase (DoD Verified)           |
+   |  - Execute verify.sh and run-evals.sh             |
+   |  - Normalize line-endings and format files        |
+   +---------------------------------------------------+
+```
+
+Never skip a phase, and never claim completion without verifiable evidence of test passing.

package/.codex/skills/genesis-harness-engineering/agents/openai.yaml

@@ -0,0 +1,7 @@
+interface:
+  display_name: "Harness Engineering Skill"
+  short_description: "Improve Codex harness reliability"
+  default_prompt: "Use $harness-engineering-skill to evolve the harness with tests first."
+policy:
+  allow_implicit_invocation: true
+

package/.codex/skills/genesis-harness-engineering/checklists/checklist.md

@@ -0,0 +1,48 @@
+# Checklist: Pre-Flight Harness Validation & Compliance
+
+Use this checklist prior to committing or declaring completion on any evolution of the **Genesis Codex Harness** operating environment.
+
+---
+
+## 🏗 Part 1: Structural & File Invariants
+
+- [ ] **Folder Structure Compliance**: Every skill has a `SKILL.md`, `agents/openai.yaml`, `templates/`, and `examples/`. Playbook-based skills also include `checklists/`, `playbooks/`, and `observability/`.
+- [ ] **Line Ending Invariants**: All script files recursively (`*.sh`) are verified to have LF line endings to avoid WSL/macOS parse failures on Windows engines.
+- [ ] **Executable Bit**: All scripts inside `scripts/` and skill-specific script folders have executable flags set (`chmod +x`).
+
+---
+
+## 📝 Part 2: Skill Frontmatter & Section Checks
+
+- [ ] **Casing Validation**: Standard skill sections match exact capitalization rules:
+  - `## Purpose`
+  - `## When to use`
+  - `## When NOT to use`
+  - `## Inputs required`
+  - `## Outputs required`
+  - `## Required tests`
+  - `## Required fixtures`
+  - `## Required contract updates`
+  - `## Required codebase map updates`
+  - `## Token saving rules`
+  - `## Acceptance criteria`
+  - `## Common mistakes`
+  - `## Recovery workflow`
+- [ ] **Metadata Alignment**: The frontmatter `name:` matches the legacy expected string checked inside `scripts/verify.sh` for directory mapping.
+
+---
+
+## 🧪 Part 3: Test Suite & Distribution Execution
+
+- [ ] **Red-State Invariant**: The test failure was demonstrated *first* before implementing the changes.
+- [ ] **Verification Smoke Pass**: `bash scripts/verify.sh` completes cleanly with exit code `0`.
+- [ ] **Regression Eval Pass**: `bash scripts/run-evals.sh` completes cleanly with exit code `0`.
+- [ ] **CLI Dry-Run Integrity**: `npm run pack:check` runs successfully and package size is correct.
+- [ ] **Uninstall Isolation**: The `uninstall.sh` script leaves no stale files behind in custom home locations.
+
+---
+
+## 💾 Part 4: Memory Invariance
+
+- [ ] **Current State Updated**: Skill counts, quality metrics, and completion statuses are synchronized in `.codebase/CURRENT_STATE.md`.
+- [ ] **Recovery Point Saved**: Exact rollback steps and verification logs are saved in `.codebase/RECOVERY_POINTS.md`.

package/.codex/skills/genesis-harness-engineering/examples/example.md

@@ -0,0 +1,57 @@
+# Example: Harness Evolution - Integrating `genesis-performance-profiling`
+
+This example demonstrates how an engineer (human or AI) safely scales the **Genesis Codex Harness** to support a new playbook-based skill called `genesis-performance-profiling`.
+
+---
+
+## Step 1: Write a Failing Verification Test (RED State)
+First, we add the new skill directory to the `skill_names` array inside `scripts/verify.sh` so that it is included in the test loop:
+```diff
+ skill_names=(
+   genesis-harness
+   genesis-new-design
+   genesis-upgrade-design
++  genesis-performance-profiling
+ )
+```
+And we add its case mapping inside `verify_one`:
+```diff
++    genesis-performance-profiling)
++      verify_skill_metadata "$skill_dir" "genesis-performance-profiling"
++      verify_playbook_skill "$skill_dir"
++      ;;
+```
+When we run `bash scripts/verify.sh`, it fails immediately because the directory `genesis-performance-profiling` does not exist yet. This confirms our validation test is active and skeptical!
+
+---
+
+## Step 2: Implement the Skill (GREEN State)
+We create the directory and populate all required files according to the playbook protocol:
+- `SKILL.md` (with all 13 standard headers and metadata).
+- `agents/openai.yaml`.
+- `templates/`, `checklists/`, `playbooks/`, `observability/`, and `examples/` directories.
+
+---
+
+## Step 3: Normalize Line Endings
+We run our CRLF-to-LF conversion command to guarantee cross-platform shell compatibility:
+```bash
+python -c "import os; [open(os.path.join(root, f), 'wb').write(content) for root, dirs, files in os.walk('.') for f in files if f.endswith('.sh') for content in [open(os.path.join(root, f), 'rb').read().replace(b'\r\n', b'\n')]]"
+```
+
+---
+
+## Step 4: Run E2E Verification
+We run all validation checks locally to confirm that the new skill is fully integrated and compliant:
+```bash
+$ bash scripts/verify.sh
+verify passed: /mnt/d/PROJECT/codex-genesis-harness/.codex/skills
+
+$ bash scripts/run-evals.sh
+evals passed
+```
+
+---
+
+## Step 5: Update Codebase Memory
+We register our evolutionary step in [.codebase/CURRENT_STATE.md](file:///.codebase/CURRENT_STATE.md) and [.codebase/RECOVERY_POINTS.md](file:///.codebase/RECOVERY_POINTS.md) before pushing our changes to version control!