review-full

STARTER_CHARACTER = 🔎

Full Review

Review the entire codebase across 9 focus areas in parallel. Each agent reads from a single repomix-packed snapshot via MCP. Output goes to per-area files in .llmtmp/review-full/.

Arguments

If the user provided a path with the invocation, treat it as the target directory relative to the repo root. Otherwise pack the whole repo.

Step 1: Resolve Target

PROJECT_ROOT=$(git rev-parse --show-toplevel)
PROJECT_ROOT=$(cd -P "$PROJECT_ROOT" && pwd -P)
TARGET_PATH="<user-provided path or empty>"
[ -z "$TARGET_PATH" ] && TARGET_PATH="$PROJECT_ROOT"

Resolve to physical absolute path and verify it stays inside the repo (no traversal, no symlink escape).

TARGET_PATH=$(cd -P "$TARGET_PATH" 2>/dev/null && pwd -P) || { echo "TARGET_PATH does not exist"; exit 1; } case "$TARGET_PATH" in "$PROJECT_ROOT"|"$PROJECT_ROOT"/*) ;; *) echo "TARGET_PATH escapes PROJECT_ROOT"; exit 1 ;; esac

TARGET_NAME=$(basename "$TARGET_PATH") [ "$TARGET_PATH" = "$PROJECT_ROOT" ] && TARGET_NAME="repo"

## Step 2: Clean and Prepare Output

```bash
OUTPUT_DIR="$PROJECT_ROOT/.llmtmp/review-full"
mkdir -p "$OUTPUT_DIR"
find "$OUTPUT_DIR" -mindepth 1 -delete

mkdir -p then find -delete rather than rm -rf. A safe-rm shim on PATH (as in some dotfiles setups) moves paths to Trash and exits non-zero on a missing path even under -f, which breaks the wipe on a first run.

Step 3: Pack Codebase

REPOMIX_FILE="$OUTPUT_DIR/repomix.xml"
npx repomix -o "$REPOMIX_FILE" --quiet --output-show-line-numbers "$TARGET_PATH"

The repomix file lives inside OUTPUT_DIR (.llmtmp/review-full/), which is gitignored and wiped at the start of every run by Step 2. No /tmp leakage and no trap needed. A trap does not work here because each bash code block in this skill runs in a separate shell process; an EXIT trap registered in Step 3 would fire as soon as Step 3's bash block ends, deleting the file before Step 4 (attach_packed_output) can read it.

Repomix reads .repomixignore from the project root automatically.

Verify the file was created. Confirm REPOMIX_FILE before proceeding.

Step 4: Attach Packed Output

Call attach_packed_output with filePath=REPOMIX_FILE. Record the returned outputId string. This is the codebase access handle the agents will use.

Do not delegate this step to a subagent. The orchestrator must own the outputId.

Step 5: Dispatch 9 Review Agents in Parallel

Issue all 9 Agent calls in a single tool block with run_in_background: false. Subagents default to running in the background, and a backgrounded fan-out delivers its results in later turns, so Step 6 would verify files that no agent has written yet.

Agent	Subagent type	Output file
Architecture & Design	reviewer-architecture	architecture.md
Correctness & Bugs	reviewer-correctness	correctness.md
Data & Information Architecture	reviewer-data	data.md
Operational Readiness	reviewer-ops	ops.md
Performance	reviewer-performance	performance.md
Code Quality	reviewer-quality	quality.md
Security	reviewer-security	security.md
SOLID Principles	reviewer-solid	solid.md
Testing	reviewer-testing	testing.md

Each review prompt contains:

1. The outputId from Step 4
2. Instruction to use read_repomix_output and grep_repomix_output for navigation
3. Instruction: "Read AGENTS.md (or CLAUDE.md) and .llmdocs/architecture.md (if present) via the repomix output for project context."
4. Instruction: "Write findings to <OUTPUT_DIR>/<area>.md. Use H2 header followed by findings or 'No findings.'"

Step 6: Wait and Verify

After all 9 agents complete, verify each file in <OUTPUT_DIR>:

1. File exists and is non-empty
2. First non-blank line is an H2 header (## <Area>)
3. Body contains at least one finding OR exactly the literal No findings.

If a file fails any of those checks, treat the agent as having silently failed and re-dispatch that one agent. Do NOT accept a stub like ## Testing\n with no body or content that omits the H2 header.

Step 7: Summarize

Produce a summary in the conversation:

# Full Review Summary

Target: <TARGET_NAME>
Files reviewed: <count from repomix>

## Findings Counts

| Area                  | High | Medium | Low |
| --------------------- | ---- | ------ | --- |
| Architecture & Design | <N>  | <N>    | <N> |
| Correctness & Bugs    | <N>  | <N>    | <N> |
| Data & Information    | <N>  | <N>    | <N> |
| Operational Readiness | <N>  | <N>    | <N> |
| Performance           | <N>  | <N>    | <N> |
| Code Quality          | <N>  | <N>    | <N> |
| Security              | <N>  | <N>    | <N> |
| SOLID Principles      | <N>  | <N>    | <N> |
| Testing               | <N>  | <N>    | <N> |

## Output Files

- `.llmtmp/review-full/architecture.md`
- `.llmtmp/review-full/correctness.md`
- `.llmtmp/review-full/data.md`
- `.llmtmp/review-full/ops.md`
- `.llmtmp/review-full/performance.md`
- `.llmtmp/review-full/quality.md`
- `.llmtmp/review-full/security.md`
- `.llmtmp/review-full/solid.md`
- `.llmtmp/review-full/testing.md`

Read individual files for detailed findings.

Do not load the file contents into the conversation. The summary is the artifact; files are persistent for follow-up.