Content
77%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The workflow is a strong, well-validated five-phase procedure with concrete thresholds, tool calls, and error-recovery loops. The main weaknesses are reference integrity — several referenced files (examples.md, templates/, src/factors/base.py) are missing from the bundle while existing reference files are unmentioned — and some duplicated guidance between the phases and Common Pitfalls.
Suggestions
Ship or fix the broken references: create templates/factor_signal_engine.py and templates/strategy_signal_engine.py (Phase 3 depends on them), add examples.md, and correct the src/factors/base.py path.
Link the existing but unmentioned bundle files (scheduled_decay_scan.md, strategy_extraction_guide.md, strategy_metrics.md) from the relevant phases or the References section so they are discoverable.
Consolidate the duplicated deduplication IC guidance (Phase 2 vs. 'Factor Deduplication' pitfall) into one location to trim tokens.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense, procedural, and assumes competence — no space is spent explaining concepts Claude already knows. Minor trimming opportunities exist: the dedup IC thresholds (0.99 / 0.90-0.99) are explained in Phase 2 and again verbatim in "Factor Deduplication", and the decay state machine transitions appear in both Phase 5 and the pitfalls. These redundancies keep it below the lean 5 anchor but well above the noticeably-verbose 3. | 4 / 5 |
Actionability | Highly concrete: named tool calls with parameters (sdm_register, generate_backtest_config with bench_type semantics), exact statistical thresholds (IC > 0.03, IR > 0.5, Sharpe > 0.5, DD < 30%), a runnable AST-validation command, and a full SignalEngine code contract. The gap: Phase 3 directs implementation "using the appropriate template from templates/" but the templates/ directory does not exist in the bundle, so that step is not fully executable as written. | 4 / 5 |
Workflow Clarity | Five phases are clearly sequenced with routing guidance, and validation checkpoints with feedback loops are explicit throughout: AST syntax validation before backtest, bench_record confirmation with "fix the config and re-run" recovery, OCR quality branches, a dedup gate, and a closing Quality Checklist. This matches the anchor for clear sequence plus explicit validation, error-recovery loops, and a checklist. | 5 / 5 |
Progressive Disclosure | In-body structure is good (sections, tool reference table, References list), but scored against the actual bundle it navigates poorly: of the referenced paths, only references/decay_thresholds.md exists — examples.md, templates/, and src/factors/base.py are missing — while three real bundle files (scheduled_decay_scan.md, strategy_extraction_guide.md, strategy_metrics.md) are never mentioned in SKILL.md and are undiscoverable. References are present but unreliable/mis-signaled, fitting the 3 anchor better than the well-organized 4 anchor and better than the 2 anchor given the body's solid internal structure. | 3 / 5 |
Total | 16 / 20 Passed |