feat: elevate codex with claude depth and shared runtime features

[mumit]

Summary

Two commits forming one logical elevation effort:

1. feat: deepen claude-dev-team parity — closes the v1.0.0 audit gaps: documents the stage-renumbering divergence, fleshes out role prompts (24 → 100-170 lines each), adds the Stage 0 safety stoplist + budget-gate config + async-checkpoint config, hardens approval-derivation.js with file locking and atomic writes, ports the audit-phases reference, and replaces the row-existence parity check with a deep content-depth check.

2. feat: elevate codex with claude depth and shared runtime features — ports claude-dev-team's prose depth AND adds three new runtime features neither framework had.

Prose depth ported from claude

• .codex/rules/pipeline.md 145 → 393 lines (review shape scoped/matrix, READ-ONLY Reviewer Rule, gate merge strategy, review round limit, stage durations, parallelism)
• .codex/rules/gates.md 74 → 280 lines (per-stage extra-field examples)
• .codex/rules/coding-principles.md 62 → 151 lines
• .codex/rules/execution-profiles.md 17 → 106 lines (full local/app-worktree/cloud model with parallelism patterns)

New narrative artifacts

• EXAMPLE.md — 218-line end-to-end pipeline walkthrough (codex-ized password-reset feature)
• CHANGELOG.md — versions v1.0.0, v1.1.0 (this), v1.2.0 (unreleased placeholder)
• CONTRIBUTING.md — local dev setup, test/lint/parity commands, PR conventions, stage-numbering note

New runtime features (neither framework had these)

Feature	Script	What it does
Budget tracking	scripts/budget.js	Honors budget.enabled in .codex/config.yml; writes pipeline/budget.md; emits stage-budget.json ESCALATE on overrun (or warns). init/update/check subcommands.
Async-checkpoint auto-pass	applyCheckpointAutoPass() in codex-team.js	Honors checkpoints.{a,b,c}.auto_pass_when config (no_warnings, all_criteria_passed); writes CHECKPOINT-AUTO-PASS: line to context. Stoplist override prevents auto-pass on security-sensitive runs.
Pipeline visualization	scripts/visualize.js	Generates a Mermaid stateDiagram-v2 of the active pipeline run, color-coded by gate status (PASS/FAIL/ESCALATE/missing). Writes pipeline/diagram.md.

Test rigor

125 → 169 tests across 16 suites. New tests:

• tests/budget.test.js — 15 tests (init, update, check escalate/warn paths, disabled-mode no-op)
• tests/checkpoints.test.js — 15 tests (each condition + null default + stoplist override)
• tests/visualize.test.js — 14 tests (empty/active/complete pipelines, valid Mermaid syntax)

Plus all earlier deepening tests (parity-check main() + mutation tests, role-prompt line-count checks, config-key validation, audit-phases reference).

Stage numbering preserved

Codex keeps its collapsed numbering (Stage 5 = pre-review with security_review_required flag). The translation table to claude's Stage 4.5a/4.5b lives in docs/parity/claude-dev-team-parity.md under "Stage Numbering Divergence".

Test plan

• npm test — all 169 pass
• npm run doctor — all PASS
• npm run parity:check — passes (deep check)
• npm run lint — passes
• npm run budget -- init — works (no-op when disabled)
• npm run visualize — writes pipeline/diagram.md with valid Mermaid
• npm run pipeline -- "Test feature" — workspace bootstraps cleanly
• npm run next — track-aware advancement still works

🤖 Generated with Claude Code