Content
50%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 body has a sound skeleton — a check-first gate, a clear decision tree, and well-signaled one-level references — but execution details undermine it: the primary raw-image code example imports a function that does not exist in the cited module, two scripts listed in the References section are missing from the bundle, one reference doc is orphaned, and the percentage-formatting rule plus ~90 lines of tool documentation bloat the file through duplication and inline bulk. Scores cluster at the midpoint across all four dimensions.
Suggestions
Fix the raw-image example to match the actual API: `from scripts.segment_cells import count_cells_watershed` then `count_cells_watershed(image, channel=0, min_area=50)` — the current `count_cells_in_image(image_path=...)` exists in neither scripts/segment_cells.py nor (with that signature) references/image_processing.md.
Reconcile the References section with the bundle: either add scripts/colony_morphometry.py and scripts/statistical_comparison.py or remove them from the list, and link references/troubleshooting.md from the body so it is discoverable.
Deduplicate the "relative proportion → percentage" guidance (currently stated in full both in the CRITICAL section and under Answer Formatting) into a single section, and move the four per-tool subsections (Cellpose, CellPainting, BioImage Archive, FPbase) into a references/ file to cut inline bulk.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient — the decision tree, patterns table, R-to-Python equivalents, and answer-formatting rules are dense and unpadded — but the "relative proportion → percentage" rule is stated twice in near-full (the CRITICAL section near the top and a subsection under Answer Formatting), and ~90 lines of per-tool API documentation (Cellpose, CellPainting, BioImage Archive, FPbase) are inlined that could be tightened or split out. It is above the 2 anchor because there is little concept-explaining filler, but below 4 due to the clear duplication and inline bulk. | 3 / 5 |
Actionability | There is concrete guidance (Phase 0 discovery code, the ratio→percentage snippet, specific tool invocations with parameters), but the showcased raw-image example — `from scripts.segment_cells import count_cells_in_image` called as `count_cells_in_image(image_path="cells.tif", channel=0, min_area=50)` — targets a function that does not exist in that module (the script defines `count_cells_watershed(image, channel=None, min_area=50, ...)`), and the keyword signature matches neither the script nor the single-argument `count_cells_in_image(image)` in references/image_processing.md. This matches the 3 anchor (concrete but incomplete / missing key details) rather than 4, since a key example fails as written. | 3 / 5 |
Workflow Clarity | A coherent gated sequence exists — RULE ZERO check-first with the explicit checkpoint "Only follow this skill's re-analysis recipe below if none of the above exist", then the pre-quantified-vs-raw-images decision tree, then Phase 0 — but Phases 1-3 collapse to a single deferred line with no inline validation checkpoints (no assumption checks before ANOVA, no group-count/N verification). This fits the 3 anchor (sequence present, checkpoints missing or implicit): above 2 because the decision tree and Rule Zero form a real gated flow, below 4 because checkpoints beyond Rule Zero are absent. | 3 / 5 |
Progressive Disclosure | References are one level deep and clearly signaled in bold (statistical_analysis.md, segmentation.md, cell_counting.md, image_processing.md all exist), but the References section lists `colony_morphometry.py` and `statistical_comparison.py` which do not exist in scripts/, `references/troubleshooting.md` exists but is never linked from the body (orphaned, undiscoverable), and the four long per-tool subsections are inlined rather than split into the bundle. This is above the 2 anchor (structure is real and mostly appropriate) but below the 4 anchor because dangling file references and an orphaned reference file are more than minor organization gaps. | 3 / 5 |
Total | 12 / 20 Passed |