T-009: Add /bp:codex-review standalone command

New slash command that invokes Codex adversarial review on demand.
Supports --base <ref> override, outputs findings in Blueprint format,
appends to impl-review-findings.md, and handles missing Codex gracefully.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

commands/codex-review.md

@@ -0,0 +1,32 @@
+---
+name: bp-codex-review
+description: "Invoke Codex adversarial review on demand — diffs current tier against worktree base and outputs findings"
+argument-hint: "[--base REF]"
+allowed-tools: ["Bash(${CLAUDE_PLUGIN_ROOT}/scripts/codex-review.sh:*)"]
+---
+
+# /bp:codex-review — Codex Adversarial Review
+
+Run an on-demand adversarial code review using the Codex CLI. This sends the current diff to Codex for analysis of bugs, security issues, logic errors, and spec violations.
+
+## Execution
+
+Run the review script, forwarding any arguments the user provided:
+
+```!
+"${CLAUDE_PLUGIN_ROOT}/scripts/codex-review.sh" $ARGUMENTS
+```
+
+## Behavior
+
+- **Default target**: diffs the current build tier's changes against the worktree base (auto-detected from upstream tracking branch, or falls back to `main`/`master`/`develop`).
+- **`--base <ref>`**: override the diff base to any git ref (branch, tag, or commit SHA).
+- Findings are printed to stdout in Blueprint's standard finding format (`F-NNN` numbered, severity `P0`–`P3`, source tagged as `codex`).
+- Findings are automatically appended to `context/impl/impl-review-findings.md`.
+- If Codex is not installed or not available, the script prints a clear message and exits gracefully — no error is thrown.
+
+## After the script completes
+
+1. **If findings were reported**: summarize them for the user — count by severity, highlight any P0/P1 items, and note the findings file path.
+2. **If no findings**: confirm a clean review to the user.
+3. **If Codex was unavailable**: inform the user that Codex CLI is required and suggest installing it (`npm install -g @openai/codex` or equivalent).

commands/help.md

@@ -14,6 +14,16 @@ description: Show Blueprint commands and usage
 /bp:inspect     →  gap analysis + peer review (the CHECK)
 ```
 
+### Quick Mode
+
+```bash
+/bp:quick <describe your feature>   # runs all 4 phases end-to-end, no stops
+/bp:quick add JWT auth --skip-inspect
+/bp:quick refactor the API layer --peer-review
+```
+
+Streamlined draft + architect (no interactive Q&A, no user gates) followed by the full build and inspect. Best for small-to-medium features where you trust the decomposition.
+
 ## Commands
 
 ### `/bp:draft` — Write Blueprints
@@ -39,14 +49,14 @@ Reads blueprints, decomposes requirements into tasks, organizes into dependency
 ### `/bp:build` — Run the Loop
 
 ```bash
-/bp:build                       # ralph loop from site
+/bp:build                       # auto-parallel build from site
 /bp:build --filter v2           # scope to v2
 /bp:build --peer-review         # add Codex (GPT-5.4) review
 /bp:build --max-iterations 30   # iteration limit
 /bp:build --peer-review --codex-model gpt-5.4-mini
 ```
 
-Auto-archives any previous cycle, then starts a Ralph Loop. Each iteration: pick unblocked task → read blueprint → implement → validate → commit.
+Auto-archives any previous cycle, then builds the site. Automatically parallelizes by grouping ready tasks into a few coherent work packets based on shared files, subsystem, and complexity. Progresses through tiers autonomously without manual intervention. If multiple build sites exist, asks which one to implement.
 
 With `--peer-review`: alternates build and review iterations, calling Codex via MCP.
 
@@ -72,10 +82,20 @@ Produces a verdict: APPROVE / REVISE / REJECT with prioritized findings.
 
 Shows tasks done/ready/blocked, progress bar, current tier, and next tasks.
 
+### `/bp:codex-review` — On-Demand Codex Review
+
+```bash
+/bp:codex-review                  # review current tier diff
+/bp:codex-review --base v1.0     # review diff against a specific ref
+```
+
+Sends the current diff to Codex for adversarial review. Outputs findings in Blueprint format and appends them to `context/impl/impl-review-findings.md`. Requires Codex CLI to be installed.
+
 ### Maintenance (optional)
 
 | Command | When |
 |---------|------|
+| `/bp:codex-review` | On-demand Codex adversarial review of current diff |
 | `/bp:gap-analysis` | After a loop — compare built vs intended |
 | `/bp:revise` | After manual code fixes — trace back to blueprints |
 | `/bp:compact-specs` | When impl tracking files exceed ~500 lines |
@@ -95,7 +115,7 @@ These still work but are superseded by the three main commands:
 | `/blueprint implement` | `/bp:build` (one task at a time vs full loop) |
 | `/blueprint spec-loop` | `/bp:build` |
 | `/blueprint peer-review-loop` | `/bp:build --peer-review` |
-| `/blueprint quick` | `/bp:draft` + `/bp:architect` + `/bp:build` |
+| `/blueprint quick` | `/bp:quick` (end-to-end with no stops) |
 
 ## Skills (reference docs)