Review Implementation

Multi-LLM implementation review for verifying code matches spec, identifying gaps, and tracking completion state.

CRITICAL: Arg Detection (Read First)

When this skill is invoked, check IMMEDIATELY if args were provided.

How to Detect Args

When invoked via Skill(skill="review-implementation", args="..."), the args string appears in your context. Look for any of these patterns:

1. Direct arg string - A line like: --models opus,gpt,gemini --categories all --scope-spec "epic cgt-22" --scope-impl "changes since start"
2. Invocation context - Check if the message that loaded this skill included an args parameter
3. User instruction - The user may have said "run /review-implementation --models opus,gpt,gemini ..."

Parsing Algorithm

If args are detected, parse them BEFORE doing anything else:

# Pseudo-code for arg parsing
args_string = [the args string you detected]

# Extract each parameter
models = extract_after("--models", args_string)  # e.g., "opus,gpt,gemini" -> ["opus", "gpt", "gemini"]
categories = extract_after("--categories", args_string)  # e.g., "all" or "completeness,quality"
scope_spec = extract_quoted_after("--scope-spec", args_string)  # e.g., "epic cgt-22"
scope_impl = extract_quoted_after("--scope-impl", args_string)  # e.g., "changes since start"
run_preflight = contains("--run-preflight-checks", args_string)  # boolean
skip_preflight = contains("--skip-preflight-checks", args_string)  # boolean

Behavior When Args Detected

If args ARE present (preflight mode):

• ✅ Skip Step 1a (scope questions) - use --scope-spec and --scope-impl
• ✅ Skip Step 1b (model/category questions) - use --models and --categories
• ✅ Skip Step 1c (detailed category questions) - categories already specified
• ✅ Skip pre-flight checks unless --run-preflight-checks is present
• ✅ Proceed directly to Step 2 (Haiku path discovery)

If args are NOT present (interactive mode):

• Ask all Step 1 questions as documented below

Verification Checkpoint

After parsing, confirm your understanding:

📋 Preflight Mode Detected
- Models: [list parsed models]
- Categories: [list parsed categories]
- Spec scope: [parsed scope-spec]
- Impl scope: [parsed scope-impl]
- Pre-flight checks: [skip/run]

Proceeding directly to Step 2...

CRITICAL: If you're unsure whether args were provided, default to INTERACTIVE mode and ask the user.

---

When to Use

• After completing implementation work against a spec
• During implementation to check progress against requirements
• Before creating a PR to verify all requirements are met
• When user asks to verify implementation matches original plan

Sequential with review-documentation: Review docs first (ensure spec is accurate), then review implementation (ensure code matches spec).

Preflight Mode (Automated Invocation)

When invoked with args, skip interactive questions and proceed directly to review.

Args Format

/review-implementation --models <models> --categories <categories> --scope-spec "<spec>" --scope-impl "<impl>" [--run-preflight-checks]

Parameters:

• --models: Comma-separated list: opus, gpt, gemini (default: opus)
• --categories: all or comma-separated: completeness,quality,scope,standards (default: all)
• --scope-spec: Spec scope description (e.g., "epic cgt-22", "specs/auth.md")
• --scope-impl: Implementation scope description (e.g., "changes in this branch", "src/auth/")
• --run-preflight-checks: Optional. Run CLI pre-flight checks for GPT/Gemini before dispatch
• --skip-preflight-checks: Optional. Explicitly skip CLI pre-flight checks (default behavior, but explicit for clarity)

Example Invocations

# Full automation - Opus only, all categories
/review-implementation --models opus --categories all --scope-spec "epic cgt-22" --scope-impl "changes since epic start"

# Multi-model with specific categories
/review-implementation --models opus,gpt --categories completeness,quality --scope-spec "specs/auth.md" --scope-impl "src/auth/"

# With pre-flight checks enabled
/review-implementation --models opus,gemini --categories all --scope-spec "epic cgt-15" --scope-impl "this branch" --run-preflight-checks

Behavior in Preflight Mode

When args are provided:

1. Skip Step 1a - Use --scope-spec and --scope-impl instead of asking
2. Skip Step 1b - Use --models and --categories instead of asking
3. Skip Step 1c - Categories already specified via --categories all or explicit list
4. Pre-flight checks - Only run if --run-preflight-checks flag present
5. Proceed directly to Step 2 (Haiku path discovery)

Use case: Automation from other skills (e.g., implementing-beads-epic) that already know the review parameters.

Workflow

flowchart TD
    start([User invokes skill])
    check_args{Args provided?}
    parse_args[Parse preflight args]
    ask_llms[Step 1a: LLMs + Scope description]
    show_cats[Step 1b: Show categories + Review All vs Customize]
    fast_track{Review All?}
    ask_cats[Step 1c: 4 multiselect questions - review categories]
    preflight_checks[Pre-flight checks IN PARALLEL]
    check_gpt[Check Codex CLI available + logged in]
    fail_gpt([Abort: tell user to fix Codex setup])
    check_gemini[Check Gemini CLI available + logged in]
    fail_gemini([Abort: tell user to fix Gemini setup])
    haiku[Step 2: Haiku gathers spec paths + implementation paths]
    build_brief[Build dynamic review brief]
    dispatch[Step 3: Dispatch reviewers IN PARALLEL]
    opus[Opus 4.6 sub-agent]
    gpt[Codex CLI GPT sub-agent]
    gemini[Gemini CLI sub-agent]
    collect[Collect results via TaskOutput]
    synthesize[Step 4: Compare, deduplicate, synthesize]
    findings([Present findings with completion matrix])
    gate[Step 5.1: Gate - wait for 'go']
    show_findings[Step 5.2: Summarize + Address All vs Customize]
    all_findings{Address All?}
    select_findings[Step 5.2b: Select specific findings]
    resolve_ambig[Step 5.3: Resolve ambiguities one-by-one]
    select_actions[Step 5.4: Select actions]
    execute[Step 5.5: Execute actions]
    summary[Step 5.6: Summary of actions taken]
    followup{Step 5.7: Issues created?}
    ask_impl[Ask: Implement now?]
    implement[Implement created issues]
    done([Done])

    start --> check_args
    check_args -->|yes - preflight mode| parse_args
    check_args -->|no - interactive mode| ask_llms
    parse_args --> haiku
    ask_llms --> show_cats
    show_cats --> fast_track
    fast_track -->|yes| preflight_checks
    fast_track -->|no| ask_cats
    ask_cats --> preflight_checks
    preflight_checks -->|if GPT selected| check_gpt
    preflight_checks -->|if Gemini selected| check_gemini
    preflight_checks -->|neither selected| haiku
    check_gpt -->|ok| haiku
    check_gpt -->|fail| fail_gpt
    check_gemini -->|ok| haiku
    check_gemini -->|fail| fail_gemini
    haiku --> build_brief
    build_brief --> dispatch
    dispatch -->|if selected| opus
    dispatch -->|if selected| gpt
    dispatch -->|if selected| gemini
    opus --> collect
    gpt --> collect
    gemini --> collect
    collect --> synthesize
    synthesize --> findings
    findings --> gate
    gate --> show_findings
    show_findings --> all_findings
    all_findings -->|yes| resolve_ambig
    all_findings -->|no| select_findings
    select_findings --> resolve_ambig
    resolve_ambig --> select_actions
    select_actions --> execute
    execute --> summary
    summary --> followup
    followup -->|yes| ask_impl
    followup -->|no| done
    ask_impl -->|yes| implement
    ask_impl -->|no| done
    implement --> done

Step 0: Pre-flight Checks (if GPT or Gemini selected)

Run pre-flight checks IN PARALLEL for all selected external CLI agents.

GPT Pre-flight (if GPT selected)

If the user selects GPT 5.3 Codex, verify Codex CLI is available before proceeding:

# Check Codex CLI is installed
command -v codex >/dev/null 2>&1 || { echo "Codex CLI not installed"; exit 1; }

# Test the actual model
codex exec -m "gpt-5.3-codex" -c reasoning_effort="high" --sandbox read-only "Respond with only: READY" 2>&1

If check fails, STOP and tell the user:

"Codex CLI check failed. Select Opus 4.6 only, or fix your Codex setup first."

Gemini Pre-flight (if Gemini selected)

If the user selects Gemini 3 Pro, verify Gemini CLI is available before proceeding:

# Check Gemini CLI is installed
command -v gemini >/dev/null 2>&1 || { echo "Gemini CLI not installed"; exit 1; }

# Test the actual model
gemini "Respond with only: READY" --model gemini-3-pro-preview -y 2>&1

If check fails, STOP and tell the user:

"Gemini CLI check failed. Select Opus 4.6 only, or fix your Gemini setup first."

Step 1: Get Scope from User (FAST)

PREFLIGHT MODE CHECK: If args were provided (see "Preflight Mode" section above), skip this entire step. Use the provided --scope-spec, --scope-impl, --models, and --categories values and proceed directly to Step 2.

CRITICAL: Do NOT explore the codebase yourself. Just collect scope from user.

Step 1a: Get Scope Descriptions

Ask the user for scope (free text):

"What should I review?

1. Spec scope - what spec/requirements to review against? (e.g., "epic cgt-22", "specs/auth.md", "tasks 1-5")
2. Implementation scope - what code to review? (e.g., "changes in this branch", "src/auth/", "all files since epic start")"

Step 1b: Configuration (Single Batched AskUserQuestion)

Batch all configuration into ONE AskUserQuestion call with 3 questions. Users can tab through efficiently:

questions: [
  {
    question: "Which models should perform this review?",
    header: "Models",
    multiSelect: true,
    options: [
      { label: "Opus 4.6 (Recommended)", description: "Claude Opus 4.6 - strong reasoning, nuanced analysis" },
      { label: "GPT 5.3 Codex", description: "OpenAI's latest via Codex CLI - different perspective" },
      { label: "Gemini 3 Pro", description: "Google's latest via Gemini CLI - third perspective" }
    ]
  },
  {
    question: "Review mode?",
    header: "Categories",
    multiSelect: false,
    options: [
      { label: "Review All (Recommended)", description: "Check all categories: completeness, quality, scope, standards" },
      { label: "Customize", description: "Select specific categories to review" }
    ]
  },
  {
    question: "Run CLI pre-flight checks for GPT/Gemini?",
    header: "Pre-flight",
    multiSelect: false,
    options: [
      { label: "Skip (Recommended)", description: "Assume CLIs work - faster startup" },
      { label: "Run checks", description: "Verify Codex/Gemini CLIs are working before dispatch" }
    ]
  }
]

This reduces LLM round-trips from multiple separate questions to 1 batched interaction.

Handling Configuration Results

Single Model Warning:
If user selects only 1 model, show a follow-up warning:

question: "Only 1 model selected. Multi-model comparison provides better coverage. Proceed?"
header: "Single Model"
multiSelect: false
options:
  - label: "Add another model (Recommended)"
    description: "Go back and select additional models for comparison"
  - label: "Continue with 1 model"
    description: "Get review from single model (no comparison)"

Review Mode Handling:

• "Review All": Skip to Step 2 with defaults:

  • Feature Completeness ✓
  • Acceptance Criteria Met ✓
  • Edge Cases Covered ✓
  • No Scope Creep ✓

• "Customize": Show the 4 multiselect questions below.

Pre-flight Handling:

• "Skip": Proceed directly to dispatch - trust that CLIs work
• "Run checks": Execute pre-flight checks (Step 0) before dispatch

Step 1c: Review Categories (if customizing - 4 multiselect questions)

Question 1: Completeness

question: "Which completeness checks should agents perform?"
header: "Completeness"
multiSelect: true
options:
  - label: "Feature Completeness (Recommended)"
    description: "Are all specified features/requirements implemented?"
  - label: "Acceptance Criteria Met (Recommended)"
    description: "Do implementations satisfy stated acceptance criteria?"
  - label: "Task Status Accuracy"
    description: "Are beads/task statuses accurate (closed = actually complete)?"

Question 2: Quality

question: "Which quality checks should agents perform?"
header: "Quality"
multiSelect: true
options:
  - label: "Edge Cases Covered (Recommended)"
    description: "Were documented edge cases handled in implementation?"
  - label: "Error Handling"
    description: "Are specified error scenarios handled correctly?"
  - label: "Test Coverage"
    description: "Are there tests for the implemented features?"

Question 3: Scope

question: "Which scope checks should agents perform?"
header: "Scope"
multiSelect: true
options:
  - label: "No Scope Creep (Recommended)"
    description: "Was ONLY what was specified implemented (no extras)?"
  - label: "Approach Alignment"
    description: "Does implementation follow specified approach (note deviations)?"
  - label: "No Missing Dependencies"
    description: "Were all prerequisite tasks completed before dependent ones?"

Question 4: Standards

question: "Which standards checks should agents perform?"
header: "Standards"
multiSelect: true
options:
  - label: "Project Conventions"
    description: "Does implementation follow CLAUDE.md and project standards?"
  - label: "API Contract Match"
    description: "Do implemented APIs match specified contracts/interfaces?"

Time budget for Step 1: Under 60 seconds total.

Step 2: Haiku Gathers Paths (FAST)

Dispatch a Haiku subagent to discover relevant paths.

Epic Disambiguation

CRITICAL: If scope description matches multiple epics (e.g., multiple "Phase 6" epics), Haiku MUST:

1. List ALL matching epics with their titles and IDs
2. Return structured output with a DISAMBIGUATION_NEEDED flag if multiple matches
3. Include epic titles to help distinguish (e.g., "Phase 6 Artifacts & Downloads" vs "Phase 6 Asset Equivalences")

If disambiguation is needed, present candidates to user:

question: "Multiple matching epics found. Which one should I review?"
header: "Epic"
multiSelect: false
options:
  [Dynamically populate with epic ID + title for each match]

Path Discovery Prompt

Task(
  subagent_type="Explore",
  model="haiku",
  prompt="Find all relevant paths for implementation review.

SPEC SCOPE: [USER'S SPEC DESCRIPTION]
IMPLEMENTATION SCOPE: [USER'S IMPLEMENTATION DESCRIPTION]

Target codebase: [PATH]

Return ONLY structured lists of paths. Do NOT read file contents.

## DISAMBIGUATION CHECK (CRITICAL)

Before returning paths, check if multiple epics/specs match the scope description.
- Run `bd list` and search for matching issues
- If the scope mentions 'Phase X' or similar, check ALL epics that mention that phase
- If multiple matches exist, return DISAMBIGUATION_NEEDED: true with list of candidates

Look for SPEC paths in:
- .beads/ directory for matching issue IDs (use bd list to find relevant IDs)
- specs/ or docs/ directories for matching markdown files
- Any paths explicitly mentioned in scope description

Look for IMPLEMENTATION paths in:
- Source directories matching implementation scope
- Test files for implemented features
- Changed files (use git diff if branch specified)

Output format:
DISAMBIGUATION_NEEDED: true/false
CANDIDATE_EPICS: (if disambiguation needed)
  - cgt-XX: [Epic Title 1]
  - cgt-YY: [Epic Title 2]
SPEC_BEADS_IDS: cgt-22, cgt-11, ...
SPEC_FILE_PATHS: /path/to/spec.md, ...
IMPL_FILE_PATHS: /path/to/impl.ts, ...
TEST_FILE_PATHS: /path/to/impl.test.ts, ...

Be thorough but fast. Paths only, no content."
)

Step 2b: Pre-Read Beads Issues

Always pre-read beads issues to include in both Opus and GPT prompts.

This speeds up the review by providing spec context upfront, reducing tool calls during review.

# For each issue ID from Haiku:
bd show cgt-XX
bd show cgt-YY
# ... etc

Store the output and include it verbatim in both sub-agent prompts under ## Beads Issue Contents (Pre-Read).

Note: GPT can still explore additional issues (dependencies, linked issues) using bd commands if needed during review.

Step 3: Dispatch Reviewers (PARALLEL)

CRITICAL: All agents MUST use run_in_background: true for true parallelism.

Special Note on Gemini: Gemini requires a TWO-STEP approach to avoid JSON serialization issues:

1. First, write the prompt to a temp file (quick sync Bash call)
2. Then, call gemini reading from the temp file (background Bash call)

See "Dispatching Gemini Sub-Agent" section below for details.

# Dispatch sequence:

# Step 1: Write Gemini prompt to temp file (sync, fast)
Bash(command="cat > /tmp/gemini_prompt.txt <<'EOF' ... EOF")

# Step 2: In ONE message, dispatch ALL reviewers with run_in_background: true:
Task(
  subagent_type="general-purpose",
  model="opus",
  run_in_background=true,    # <-- REQUIRED
  prompt="..."
)

Bash(
  command="codex exec ...",
  run_in_background=true,    # <-- REQUIRED
  timeout=900000
)

Bash(
  command="gemini \"$(cat /tmp/gemini_prompt.txt)\" ...",
  run_in_background=true,    # <-- REQUIRED
  timeout=900000
)

If you forget run_in_background: true on any call, they will NOT run in parallel.

Sub-Agent Brief (DYNAMIC)

Build based on user selections from Step 1b.

Template:

You are reviewing IMPLEMENTATION against a SPECIFICATION.

Your job is to verify the code matches what was specified, identify gaps, and assess completion state.

## IMPORTANT: Review Only - No Changes

**DO NOT modify any files during this review phase.**

This is a READ-ONLY review for context gathering and gap identification. Your role is to:
- Read and analyze code
- Compare implementation against spec
- Identify gaps and issues
- Report findings

Any actual code changes will be made in a later phase after findings are reviewed.
Do NOT use Edit, Write, or any file modification tools.

## Review Mode: Approach-Aware

- If the outcome matches the spec, implementation is correct even if approach differs
- Note approach deviations as INFORMATIONAL, not failures
- Only flag as issues if the OUTCOME doesn't match spec

## Verification Requirement

**CRITICAL: Do NOT make assumptions about how the codebase works. VERIFY by reading code.**

Before stating how something works (e.g., "the CLI is workspace-agnostic"):
1. Read the actual code that implements it
2. Quote specific lines/functions that prove your assertion
3. If you cannot verify by reading code, mark as UNVERIFIED and flag for human review

Common assumption mistakes to avoid:
- Assuming a component hasn't changed since last review
- Inferring behavior from naming conventions without reading implementation
- Stating capabilities without checking the actual code

## Later-Phase Awareness

**CRITICAL: Before recommending improvements, check if they're already scoped for later.**

When reviewing against an issue that's part of a larger epic or brief:
1. Examine the full epic/parent issue to see all planned phases
2. Check sibling issues for work already scoped in later phases
3. **Do NOT recommend bringing forward work that's already properly scoped for a later phase**

If an improvement IS already scoped later, note it as:
> "Future-scoped: [description] is covered by [issue-id] in a later phase"

Only recommend additions/improvements if they:
- Are genuinely missing from all phases
- Are blocking current phase completion
- Represent a gap in the overall plan

## DO CHECK - Review these categories:

[Include ONLY categories user selected]

### Feature Completeness (if selected)
- Is each specified feature/requirement implemented?
- Create a checklist: requirement → implemented? → where?

### Acceptance Criteria Met (if selected)
- For each acceptance criterion, is it satisfied?
- Quote the criterion and explain how implementation meets it (or doesn't)

### Task Status Accuracy (if selected)
- Are closed tasks actually complete?
- Are open tasks actually incomplete?
- Flag any status mismatches

### Edge Cases Covered (if selected)
- For each documented edge case, is it handled?
- Where in the code is it handled?

### Error Handling (if selected)
- Are specified error scenarios handled?
- Do error messages match spec?

### Test Coverage (if selected)
- Are there tests for implemented features?
- Do tests cover specified scenarios?

### No Scope Creep (if selected)
- Is there code that wasn't specified?
- Flag additions that weren't in spec (INFORMATIONAL unless problematic)

### Approach Alignment (if selected)
- Does implementation follow specified approach?
- Note deviations as INFORMATIONAL with brief explanation

### No Missing Dependencies (if selected)
- Were prerequisite tasks completed?
- Are there incomplete dependencies?

### Project Conventions (if selected)
- Does code follow CLAUDE.md conventions?
- Naming, structure, patterns match project style?

### API Contract Match (if selected)
- Do implemented APIs match specified interfaces?
- Request/response shapes, error codes, etc.

## DO NOT CHECK - Skip these categories:

[List categories user did NOT select]

## Output Format

### Completion Matrix

| Requirement | Status | Location | Notes |
|-------------|--------|----------|-------|
| [From spec] | DONE/PARTIAL/MISSING/DEVIATED | file:line | [Details] |

### Issues (Risk-Weighted)

For each issue:
- **[SEVERITY] Issue Title**
  - Requirement: What was specified
  - Actual: What was implemented (or not)
  - Location: File and line
  - Recommendation: How to resolve

### Summary Stats
- Requirements: X total, Y complete, Z partial, W missing
- Scope creep items: N (if checking)
- Approach deviations: N (if checking)

Dispatching Opus Sub-Agent

Task(
  subagent_type="general-purpose",
  model="opus",
  run_in_background=true,
  prompt="[Sub-agent brief above]

SPECIFICATION to review against:
- Spec files: [SPEC_FILE_PATHS from Haiku]

## Beads Issue Contents (Pre-Read)

[PASTE VERBATIM OUTPUT FROM `bd show` FOR EACH ISSUE HERE]

IMPLEMENTATION to review:
- Implementation files: [IMPL_FILE_PATHS from Haiku]
- Test files: [TEST_FILE_PATHS from Haiku]

Target codebase: /path/to/project

## How to explore additional beads issues

The main spec issues are provided above. If you need to explore dependencies or linked issues:
- bd show <id> - View issue details
- bd dep show <id> - View issue dependencies
- bd list --status=open - List open issues

Use Read tool for all source files.
Compare spec requirements against actual implementation.

REMINDER: This is a READ-ONLY review. Do NOT modify any files - no Edit or Write tool calls."
)

Dispatching GPT Sub-Agent via Codex CLI

Pre-read beads issues are included in the prompt for faster review. GPT can explore additional issues if needed.

# Bash tool call with run_in_background: true, timeout: 900000 (15 minutes)

codex exec -m "gpt-5.3-codex" -c reasoning_effort="high" --sandbox workspace-write "$(cat <<'PROMPT'
[Sub-agent brief above]

SPECIFICATION to review against:
- Spec files: [SPEC_FILE_PATHS]

## Beads Issue Contents (Pre-Read)

[PASTE VERBATIM OUTPUT FROM `bd show` FOR EACH ISSUE HERE]

IMPLEMENTATION to review:
- Implementation files: [IMPL_FILE_PATHS]
- Test files: [TEST_FILE_PATHS]

## How to explore additional beads issues

The main spec issues are provided above. If you need to explore dependencies or linked issues:
- bd show <id> - View issue details
- bd dep show <id> - View issue dependencies
- bd list --status=open - List open issues

## Shell Command Best Practices

**CRITICAL: Path quoting**
Always wrap file paths containing special characters in single quotes:
- Parentheses: `(dashboard)`
- Brackets: `[slug]`
- Spaces

Example:
```bash
# WRONG - will fail
sed -n '1,100p' apps/web/app/(dashboard)/workspace/[slug]/page.tsx

# CORRECT
sed -n '1,100p' 'apps/web/app/(dashboard)/workspace/[slug]/page.tsx'

Avoid login shell flag
When running shell commands, prefer non-login shell:

/bin/zsh -c 'command'  # Preferred - non-login shell

Avoid login shell (-l flag) as it may trigger profile errors in sandbox environments.

Time Management

This review may take 10-15 minutes. Take your time to:

1. Read all relevant files thoroughly
2. Cross-reference spec against implementation
3. Build a complete completion matrix

Don't rush - thoroughness is more important than speed.

Review process

1. Read the beads issue contents provided above
2. If needed, explore dependencies or linked issues using bd commands
3. Read spec markdown files and implementation files
4. Compare: does implementation satisfy spec?
5. Return findings in specified output format

REMINDER: This is a READ-ONLY review. Do NOT modify any files.
PROMPT
)"

### Dispatching Gemini Sub-Agent via Gemini CLI

**CRITICAL: Use the TWO-STEP temp file approach to avoid JSON serialization issues with complex prompts.**

**WHY:** When the executing LLM constructs a Bash tool call containing a long HEREDOC with markdown, code blocks, and special characters, the JSON serialization can break (resulting in "Invalid tool parameters"). Using a temp file separates the complex prompt content from the shell command.

**Pre-read beads issues are included in the prompt for faster review. Gemini can explore additional issues if needed.**

**IMPORTANT: Gemini runs without sandbox (docker/podman not required). The prompt explicitly instructs Gemini NOT to modify files during review phase.**

**Step 1: Write prompt to temp file (first Bash call)**

```bash
# Bash tool call parameters:
#   command: "cat > /tmp/gemini_review_prompt.txt <<'GEMINI_PROMPT' ... "
#   run_in_background: false    <-- This is fast, no need for background
#   timeout: 30000              <-- 30 seconds is plenty

cat > /tmp/gemini_review_prompt.txt <<'GEMINI_PROMPT'
[Sub-agent brief above]

SPECIFICATION to review against:
- Spec files: [SPEC_FILE_PATHS]

## Beads Issue Contents (Pre-Read)

[PASTE VERBATIM OUTPUT FROM `bd show` FOR EACH ISSUE HERE]

IMPLEMENTATION to review:
- Implementation files: [IMPL_FILE_PATHS]
- Test files: [TEST_FILE_PATHS]

## How to explore additional beads issues

The main spec issues are provided above. If you need to explore dependencies or linked issues:
- bd show <id> - View issue details
- bd dep show <id> - View issue dependencies
- bd list --status=open - List open issues

## Shell Command Best Practices

**CRITICAL: Path quoting**
Always wrap file paths containing special characters in single quotes:
- Parentheses: `(dashboard)`
- Brackets: `[slug]`
- Spaces

Example:
```bash
# WRONG - will fail
cat apps/web/app/(dashboard)/workspace/[slug]/page.tsx

# CORRECT
cat 'apps/web/app/(dashboard)/workspace/[slug]/page.tsx'

Time Management

This review may take 10-15 minutes. Take your time to:

1. Read all relevant files thoroughly
2. Cross-reference spec against implementation
3. Build a complete completion matrix

Don't rush - thoroughness is more important than speed.

Review process

1. Read the beads issue contents provided above
2. If needed, explore dependencies or linked issues using bd commands
3. Read spec markdown files and implementation files
4. Compare: does implementation satisfy spec?
5. Return findings in specified output format

REMINDER: This is a READ-ONLY review. Do NOT modify any files.
GEMINI_PROMPT

**Step 2: Call Gemini reading from temp file (second Bash call, IN PARALLEL with Opus/GPT)**

```bash
# Bash tool call parameters:
#   command: "gemini \"$(cat /tmp/gemini_review_prompt.txt)\" ..."
#   run_in_background: true     <-- REQUIRED FOR PARALLEL EXECUTION
#   timeout: 900000             <-- 15 minutes

gemini "$(cat /tmp/gemini_review_prompt.txt)" --model gemini-3-pro-preview -y && rm /tmp/gemini_review_prompt.txt

Notes on Gemini CLI:

• Model: gemini-3-pro-preview
• Runs without sandbox (prompt prohibits file modifications)
• Timeout: Set Bash timeout to 15 minutes (timeout: 900000 ms)
• Background: Use run_in_background: true on the Bash call - THIS IS CRITICAL
• The && rm cleanup runs after Gemini completes

Collecting Results

CRITICAL: Call ALL TaskOutput calls IN PARALLEL in ONE message.

If you call them sequentially, the main thread blocks on each one, wasting time. Call them all in a single message:

# In ONE message, call all TaskOutput tools in parallel:
TaskOutput(task_id="<opus_task_id>", block=true, timeout=900000)
TaskOutput(task_id="<codex_task_id>", block=true, timeout=900000)
TaskOutput(task_id="<gemini_task_id>", block=true, timeout=900000)

Timeouts: Use 900000ms (15 minutes) for all agents. Reviews take time - don't give up early.

Do NOT timeout prematurely. When waiting for results:

• Use timeout: 900000 (15 minutes) for all agents
• If one agent times out but others succeed, note this in the output and proceed
• Do NOT poll repeatedly - call TaskOutput ONCE with blocking mode and let it wait

If you get a timeout and suspect the agent is still running:

1. Wait an additional 2-3 minutes
2. Try TaskOutput again with block: false to check status
3. Only proceed to synthesis when you have actual results (or confirmed failure)

Handling Failures (Rate Limits, Timeouts, Errors)

Known Sandbox Errors (Ignorable):

These stderr messages can be safely ignored - they don't affect command execution:

• /opt/homebrew/Library/Homebrew/help.sh: cannot create temp file - Homebrew shell integration failing in sandbox
• Similar brew/nvm/pyenv/rbenv profile errors
• Any "Operation not permitted" errors from shell profile scripts

Only treat as actual failures:

• Command exit codes ≠ 0
• Missing expected output
• Actual tool/permission errors (not profile initialization errors)

If GPT fails (rate limit, timeout, or error):

1. Notify the user explicitly:

   "GPT review failed due to [rate limit/timeout/error]. Proceeding with remaining agents."

2. Continue with other agent results - don't abort the entire review

3. Note in final output:

   • Add to Review Configuration: GPT: Failed ([reason])
   • Adjust Agent Comparison section to show only available agents
   • Note reduced cross-validation if only one agent succeeded

If Gemini fails (rate limit, timeout, or error):

1. Notify the user explicitly:

   "Gemini review failed due to [rate limit/timeout/error]. Proceeding with remaining agents."

2. Continue with other agent results - don't abort the entire review

3. Note in final output:

   • Add to Review Configuration: Gemini: Failed ([reason])
   • Adjust Agent Comparison section to show only available agents

If Opus fails: Ask user if they want to retry or proceed with remaining agents.

If all selected agents fail: Report the errors and ask user how to proceed.

Transition to Synthesis

Once all TaskOutput calls return (or timeout/fail):

1. Do NOT do additional exploration - use only what agents returned
2. Do NOT supplement with your own reads - synthesis uses agent outputs only
3. Proceed IMMEDIATELY to Step 4 - don't add extra investigation
4. If results are incomplete, note gaps in the synthesis rather than trying to fill them

Step 4: Synthesize Results

Single-Agent Mode

1. Present findings directly
2. Add your own analysis
3. Note that cross-validation wasn't performed

Multi-Agent Mode

4.1 Merge Completion Matrices

Create unified matrix showing all agents' assessments (include only columns for agents that were selected/succeeded):

Requirement	Opus	GPT	Gemini	Consensus	Location
[Req]	DONE	DONE	DONE	DONE	file:line
[Req]	PARTIAL	DONE	PARTIAL	REVIEW	file:line

Table Formatting Guidelines:

• Keep cell content concise - under 30 characters per cell where possible
• Use status shorthand - DONE, PARTIAL, MISSING, DEVIATED (not full descriptions)
• If requirement text is long, abbreviate in table and reference full list above
• Test table renders correctly before presenting to user
• Use consistent column widths - don't let one cell dominate

4.2 Deduplicate Issues

Merge issues, noting agreement/disagreement.

4.3 Reasoning Section

For disagreements, explain which assessment is correct and why.

4.4 Remaining Ambiguities

List unresolved questions needing human input.

Output Format

# Implementation Review: [Spec/Epic Name]

## Review Configuration
- **LLMs Used:** [List agents used, e.g., "Opus + GPT + Gemini" / "Opus only" / "Opus + Gemini"]
- **Spec Reviewed:** [List of specs/issues]
- **Implementation Reviewed:** [List of impl files]
- **Categories Checked:** [List]
- **Categories Skipped:** [List]

## Completion Matrix

| Requirement | Status | Location | Notes |
|-------------|--------|----------|-------|
| ... | ... | ... | ... |

## Summary Stats
- **Total Requirements:** X
- **Complete:** Y (Z%)
- **Partial:** A
- **Missing:** B
- **Scope Creep Items:** C
- **Approach Deviations:** D (informational)

## Issues (Risk-Weighted)

### CRITICAL
- [Issues that block completion]

### HIGH
- [Significant gaps]

### MEDIUM
- [Minor gaps or quality issues]

### LOW / INFORMATIONAL
- [Scope creep, approach deviations]

## Agent Comparison (if multi-agent)

| # | Requirement | Opus | GPT | Gemini | Agree? |
|---|-------------|------|-----|--------|--------|
| ... |

(Include only columns for agents that were selected/succeeded)

## Reasoning (if multi-agent)

[Explain disagreements and final assessment]

## Remaining Ambiguities

- [Unresolved questions]
- [Items needing human input]

Step 5: Next Steps (Interactive)

5.1 Gate: Let User Digest Findings

"Review complete. Say 'go' when you're ready to discuss next steps."

5.2 Select Findings to Address (fast-track or customize)

First, summarize the findings:

"Found X issues: Y critical, Z high, W medium, V low/informational."

Then ask (single choice):

question: "How would you like to proceed?"
header: "Findings"
multiSelect: false
options:
  - label: "Address All Issues (Recommended)"
    description: "Work through all identified issues"
  - label: "Customize"
    description: "Select specific issues to address"

If "Address All": Proceed to Step 5.3 with all findings selected.

If "Customize": Show the multiselect of individual findings:

question: "Which findings do you want to address?"
header: "Select Issues"
multiSelect: true
options:
  [Dynamically populate from identified issues]

5.3 Resolve Ambiguities

For each ambiguity in selected findings, present one at a time with options.

5.4 Select Actions

question: "What actions do you want to take?"
header: "Actions"
multiSelect: true
options:
  - label: "Create beads issues for gaps"
    description: "Track missing/incomplete items as new issues"
  - label: "Update spec to match implementation"
    description: "If implementation is intentionally different, update spec"
  - label: "Save review to markdown"
    description: "Save full review to docs/reviews/[name].md"
  - label: "Mark tasks complete in beads"
    description: "Close issues verified as complete"
  - label: "Fix implementation gaps"
    description: "Implement missing requirements"

5.5 Execute Selected Actions

If "Create beads issues for gaps" selected:

Ask structure preference (epic with tasks vs individual issues), then create via bd create.

Best practices:

• Link new issues to parent issue with --parent at create time, or bd dep add <new-id> <parent-id> --type parent-child
• Run bd sync after creating issues to persist to sync branch

If "Update spec to match implementation" selected:

Update markdown files or beads issues via bd update to reflect intentional deviations.

If "Save review to markdown" selected:

Ask for name, save to docs/reviews/[name].md.

If "Mark tasks complete in beads" selected:

For each verified-complete task, run bd update <id> --status=closed.

If "Fix implementation gaps" selected:

Work through missing requirements, implementing each one.

5.5b Verify Applied Changes

After applying any changes (spec updates, implementation fixes, beads updates), verify they were applied correctly:

1. Re-read each modified file to confirm changes applied as intended
2. Check for syntax issues - ensure code compiles, markdown renders, tests pass
3. Summarize changes - briefly note what changed in each file for user awareness

Example verification:

Verifying changes...
- packages/auth/service.ts: ✓ Added missing validation (lines 45-52)
- specs/auth.md: ✓ Updated to reflect implementation approach
- .beads/cgt-123.md: ✓ Status updated to closed

If any change failed to apply, note it in the summary and suggest manual intervention.

5.6 Summary

## Actions Taken

- **Ambiguities resolved:** X of Y
- **Beads issues created:** [list IDs]
- **Spec updates applied:** X
- **Review saved to:** [path]
- **Tasks marked complete:** [list IDs]
- **Implementation fixes:** X requirements addressed
- **Deferred:** [list any skipped items]

5.7 Follow-Up: Implement Created Issues

If beads issues were created in Step 5.5, ask if user wants to implement them now:

question: "Beads issues created for gaps. Want to implement them now?"
header: "Implement"
multiSelect: false
options:
  - label: "Yes - implement now"
    description: "Work through the newly created issues immediately"
  - label: "No - defer to later"
    description: "Issues are tracked, implement in a future session"

If "Yes": Work through each created issue, implementing the requirements. After each implementation:

• Mark the issue as closed via bd update <id> --status=closed
• Run bd sync to persist

If "No": End the skill. Issues remain open for future work.

Execution Checklist

Use TodoWrite to track each step:

• Asked user which LLMs + spec scope + implementation scope (Step 1a)
• Showed categories and asked Review All vs Customize (Step 1b)
• If Customize: asked user which review categories (Step 1c - 4 questions)
• Pre-flight checks passed IN PARALLEL (if GPT or Gemini selected)
• Dispatched Haiku to gather spec + implementation paths
• Received path lists from Haiku
• If disambiguation needed: asked user to select correct epic
• Pre-read all beads issues with bd show for inclusion in agent prompts
• Built dynamic sub-agent brief
• Dispatched ALL selected reviewers with run_in_background: true in ONE message
• Collected results using TaskOutput
• If any agent failed: notified user and proceeded with remaining agents
• Created completion matrix
• Created comparison table (multi-agent only)
• Wrote reasoning for disagreements
• Listed remaining ambiguities
• Produced final summary with stats
• Gated: waited for user "go"
• Summarized findings and asked Address All vs Customize (Step 5.2)
• If Customize: asked user which specific findings to address
• Resolved ambiguities one-by-one
• Asked user which actions to take
• Executed selected actions (including bd sync after beads creation)
• Verified applied changes (re-read files, check syntax, summarize changes)
• Presented final summary of actions taken
• If issues created: asked user if they want to implement now (Step 5.7)
• If implementing: worked through issues, marked closed, synced

Common Mistakes

Mistake	Correction
Exploring codebase yourself	Haiku gathers paths - you pass scope descriptions
Missing run_in_background	BOTH Task and Bash calls need run_in_background: true
Sequential dispatch	ONE message with multiple tool calls for parallelism
Failing approach deviations	Approach deviations are INFORMATIONAL if outcome matches
Binary pass/fail	Use completion matrix with DONE/PARTIAL/MISSING/DEVIATED
Ignoring WIP state	Report current state accurately, don't require 100% completion
Skipping synthesis	Raw agent output isn't the deliverable - synthesize
Ending after beads creation	If issues created, MUST ask if user wants to implement now
Forgetting bd sync	Always sync after creating/updating beads issues
Recommending later-phase work	Check epic/sibling issues before recommending - don't bring forward work already scoped later
Skipping Review All option	Always offer fast-track first - only show detailed questions if user picks Customize
Not disambiguating epics	If multiple epics match (e.g., multiple "Phase 6"), ask user which one
Not pre-reading beads issues	Always pre-read with bd show and include in all agent prompts
Making technical assumptions	Don't assume how code works - READ it and quote specific lines
Silent rate limit fallback	If any agent fails, NOTIFY user explicitly before proceeding with remaining agents
Modifying files during review	Review phase is READ-ONLY - no Edit/Write calls, changes come later
Premature timeout	Use 900000ms (15 min) timeout for all agents - don't give up after 2-5 minutes
Unquoted special paths	Always single-quote paths with (), [], or spaces
Treating sandbox stderr as errors	Homebrew/profile errors in stderr are ignorable - check exit codes instead
Sequential pre-flight checks	Run GPT and Gemini pre-flight checks IN PARALLEL, not sequentially
Gemini file modifications	Gemini runs unsandboxed - prompt explicitly prohibits file modifications during review
Inline Gemini HEREDOC	Use TWO-STEP temp file approach - write prompt to file first, then call gemini
Sequential TaskOutput calls	Call ALL TaskOutput in ONE message for parallel collection
Extra exploration after collection	Proceed IMMEDIATELY to synthesis - don't supplement with your own reads
Skipping change verification	After applying changes, re-read files and summarize what changed
Table cell overflow	Keep completion matrix cells under 30 chars - use status shorthand