| name | plan-reviewer |
| description | Review LLM-generated plans for autonomous agent executability using 8-dimension rubric. Triggers: "review plan", "compare plans", "plan quality", "meta-review". |
Plan Reviewer
Quick Start
target_file: plans/my-plan.md
review_mode: FULL
Output: reviews/plan-reviews/<plan-name>-<model>-<date>.md
Purpose
Review LLM-generated plans for autonomous agent executability using an 8-dimension rubric optimized for Priority 1 compliance (Agent Understanding).
Plans are scored on whether autonomous agents can execute them without judgment calls or clarification requests.
Use this skill when
- Review a plan file for agent executability
- Compare multiple plans for the same task (choose winner)
- Conduct meta-review to analyze consistency across reviews
- Verify plan meets Priority 1 compliance before execution
What NOT to Use This For
- General documentation review → Use
doc-reviewer instead
- Rule file validation → Use
rule-reviewer instead
- Code review → Use standard code review tools
- Task execution → This skill evaluates plans, it doesn't execute them
Inputs
Required:
- review_mode:
FULL | COMPARISON | META-REVIEW | DELTA
- target_file(s): Path(s) to plan file(s) - varies by mode
Mode-specific:
| Mode | Required Input | Description |
|---|
| FULL | target_file | Single plan file path |
| COMPARISON | target_files | List of plan file paths |
| META-REVIEW | review_files | List of review file paths |
| DELTA | target_file + baseline_review | Current plan + prior review |
Optional:
output_root: Output directory (default: reviews/)
execution_mode: parallel (default, 8 sub-agents) or sequential
timing_enabled: Enable execution timing (default: true — v2.5.0 universal default; set false to opt out with not-requested row)
overwrite: Overwrite existing files (default: false)
Review Modes
| Mode | Purpose | Output Location |
|---|
| FULL | Comprehensive single-plan review | {root}/plan-reviews/<name>-<model>-<date>.md |
| COMPARISON | Rank multiple plans, declare winner | {root}/summaries/_comparison-*.md |
| META-REVIEW | Analyze consistency across reviews | {root}/summaries/_meta-*.md |
| DELTA | Track issue resolution from baseline | {root}/plan-reviews/<name>-delta-*.md |
See: examples/ for complete walkthroughs of each mode.
Review Rubric
See: rubrics/SCORING.md for complete scoring formula and verdict thresholds.
Quick Reference:
- Critical (75 pts): Executability (20), Completeness (20), Success Criteria (20), Scope (15)
- Standard (25 pts): Dependencies (10), Decomposition (5), Context (5), Risk Awareness (5)
- Verdicts: 90-100 EXCELLENT, 80-89 GOOD, 60-79 NEEDS_WORK, 40-59 POOR, <40 INADEQUATE
Detailed rubrics: rubrics/[dimension].md
Workflow
-
Validate inputs → workflows/input-validation.md
-
Collect ALL parameters (use ask_user_question) → workflows/parameter-collection.md
- MANDATORY: Prompt for ALL parameters (required AND optional) in a single question batch
- Do NOT silently apply defaults for optional parameters
- User must explicitly confirm each setting
-
Slug model name → workflows/model-slugging.md
-
[Optional] Start timing if timing_enabled: true (see Timing section below for full Quick Reference)
4a. (If timing_enabled: true) Bracket EACH of the 8 dimensions with a checkpoint pair.
Required checkpoint names (exact spelling; dim_<name>_start / dim_<name>_end):
- dim_executability_start / dim_executability_end
- dim_completeness_start / dim_completeness_end
- dim_success_criteria_start / dim_success_criteria_end
- dim_scope_start / dim_scope_end
- dim_dependencies_start / dim_dependencies_end
- dim_decomposition_start / dim_decomposition_end
- dim_context_start / dim_context_end
- dim_risk_awareness_start / dim_risk_awareness_end
On timing-end, pass --auto-dimension-timings (preferred) or assemble
--dimension-timings JSON manually. FAILURE MODE: the Per-Dimension Timing
section will be silently absent from the review; the post-write Gate 7
in workflows/file-write.md will REJECT the review.
-
Execute review per execution_mode:
parallel: 8 sub-agents → workflows/parallel-execution.md
sequential: Single agent → workflows/review-execution.md
-
[Optional] End timing and embed metadata. Include --auto-dimension-timings when timing_enabled: true.
-
Write output → workflows/file-write.md (Gate 7 enforces Per-Dimension Timing presence)
-
Handle errors → workflows/error-handling.md
CRITICAL: Default to parallel execution. Do NOT silently use sequential.
Timing (When timing_enabled: true)
You capture, skill-timing validates and formats. Passing no dimension data
results in a silently-omitted Per-Dimension Timing section unless Gate 7
catches it. Do not rely on skill-timing to "handle all" capture — it only
validates and formats what you provide.
Quick Reference (copy-paste)
PYTHON=$(bash skills/skill-timing/scripts/find_python.sh)
$PYTHON skills/skill-timing/scripts/skill_timing.py start \
--skill plan-reviewer --target plans/my-plan.md --model {{model}} --mode FULL
$PYTHON skills/skill-timing/scripts/skill_timing.py checkpoint \
--run-id {{_timing_run_id}} --name skill_loaded
$PYTHON skills/skill-timing/scripts/skill_timing.py checkpoint \
--run-id {{_timing_run_id}} --name dim_{name}_start
$PYTHON skills/skill-timing/scripts/skill_timing.py checkpoint \
--run-id {{_timing_run_id}} --name dim_{name}_end
$PYTHON skills/skill-timing/scripts/skill_timing.py checkpoint \
--run-id {{_timing_run_id}} --name review_complete
$PYTHON skills/skill-timing/scripts/skill_timing.py end \
--run-id {{_timing_run_id}} \
--output-file reviews/plan-reviews/{{plan}}-{{model}}-{{date}}.md \
--skill plan-reviewer --format markdown --auto-dimension-timings
Verify: end stdout must contain PER_DIMENSION_STATUS=present or
PER_DIMENSION_STATUS=derived. missing triggers Gate 7 rejection.
Common Timing Mistakes (Anti-Patterns)
-
Fabricated epochs — DO NOT invent timestamps.
- WRONG: Hand-writing
--dimension-timings '[{"dimension":"scope","start_epoch":1700000000,...}]' with guessed numbers.
- Correct: Emit real
checkpoint calls and pass --auto-dimension-timings.
-
Missing required fields in manual --dimension-timings.
- WRONG:
[{"dimension":"scope","duration_seconds":1.2}] (no epochs, no mode).
- Correct: See
skills/skill-timing/schemas/ for required fields (dimension, start_epoch, end_epoch, duration_seconds, mode).
-
Ignoring VALIDATION ERROR from skill-timing.
- WRONG: Seeing
VALIDATION ERROR: dimension_timings rejected and proceeding anyway.
- Correct: Treat as a hard failure; re-run
end with corrected data.
-
Omitting --auto-dimension-timings / --dimension-timings entirely.
- WRONG: Capturing
dim_*_start/end checkpoints but calling end without either flag.
- Correct: Always pass
--auto-dimension-timings when timing_enabled: true.
Determinism
See: workflows/determinism.md for complete determinism requirements.
Key rules:
- Batch load ALL 9 rubric files BEFORE reading plan
- Create ALL 8 worksheets BEFORE scoring
- Read plan line 1 to END (no skipping)
- Include worksheets in output for audit trail
- Expected variance: ±2 points max
Parallel Execution
See: workflows/parallel-specs.md for timeout handling, aggregation schema, edge cases, and rollback procedures.
Hard Requirements
- Do NOT ask user to copy/paste review
- Do NOT print entire review if file write succeeds
- Count blocking issues accurately
- Apply weighted scoring formula correctly
- If file write fails: Print
OUTPUT_FILE: <path> then full review
Examples
examples/full-review.md - FULL mode walkthrough
examples/comparison-review.md - COMPARISON mode
examples/meta-review.md - META-REVIEW mode
examples/edge-cases.md - Edge case handling
Related Skills
- rule-creator - Create rules (similar executability criteria)
- doc-reviewer - Review documentation (complementary)
- skill-timing (≥ v1.5.0) - Provides
--auto-dimension-timings used by Step 4a and Gate 7.
References
rules/000-global-core.md - Priority hierarchy definition
rules/002h-claude-code-skills.md - Skill best practices
Version History
See CHANGELOG.md.