Leaf Measure

leaf-measure

Use this skill to run the shared leaf-measure engine, not to re-implement the method in the prompt.

This skill supports two host patterns:

• repo-local: the current workspace is the leaf-measure repository
• standalone installed skill: the skill lives under $CODEX_HOME/skills/leaf-measure and uses scripts/setup_and_analyze.py to clone or update the shared repo cache under $CODEX_HOME/vendor/leaf-measure, then sync the installed skill to the latest canonical bundle

Workflow

1. Confirm the target image folder.
2. If the user did not specify a mode, explain Full image vs Thumbnails using references/mode-selection.md and ask them to choose.
3. Before analysis, make sure the runtime exists:
   • repo-local: prefer .\scripts\bootstrap.ps1 on Windows because it installs the current Python dependencies, downloads Fiji if missing, and fetches the public Figshare assets if missing
   • standalone installed skill: run python scripts/setup_and_analyze.py analyze ...; that helper clones or updates the shared repo cache, refreshes the installed skill from the latest canonical skills/leaf-measure/, runs doctor, and bootstraps the runtime on first use when needed
   • if only the upstream package is missing in a repo-local workspace, run python -m engine.cli fetch-assets
   • if you must pass Fiji explicitly, --fiji may be either a Fiji directory or a launcher path such as ImageJ-win64.exe
4. Run the shared CLI from the repository root:

python -m engine.cli analyze --input "<folder>" --output "<run-dir>" --mode full

or

python -m engine.cli analyze --input "<folder>" --output "<run-dir>" --mode thumbnails

For a standalone installed skill, use the bundled helper instead of assuming the current directory is the repo:

python scripts/setup_and_analyze.py analyze --input "<folder>" --output "<run-dir>" --mode full

or

python scripts/setup_and_analyze.py analyze --input "<folder>" --output "<run-dir>" --mode thumbnails

If the user says the installed skill still behaves like an older version, refresh it explicitly:

python scripts/setup_and_analyze.py self-update

If you are inside the repository itself and want to refresh the global installation from the canonical source, run:

python -m engine.cli update-skill

5. Read manifest.json, results.csv, run_summary.md, and the output folders before answering.
6. Explain:
   • what was measured
   • which mode was used
   • that results.csv is the user-facing table and results_fameles_particles_raw.csv preserves the original Fiji particle-level table when present
   • for Full image, whether the run had to correct a hole-dominated mask polarity before writing the user-facing table
   • for Full image, whether a conservative crop-rescue pass was needed to recover missed large leaf objects by rerunning the original Full image macro on candidate crop regions
   • that outputs are in pixels by default
   • whether DPI metadata was found
   • whether automatic repair was triggered for a mask artifact
   • whether leaf-measure internally staged non-ASCII filenames and then restored the original names in the delivered outputs
   • that binary and outline outputs should be visually reviewed

References

• Mode choice: references/mode-selection.md
• Trait definitions: references/trait-definitions.md
• Artifact handling: references/artifact-handling.md

Boundaries

• Do not silently convert pixel outputs into physical units.
• Do not claim the method worked if results.csv or output images are missing.
• Treat the shared CLI as the source of truth.
• If the standalone helper cannot make the runtime ready, surface the bootstrap or doctor failure clearly instead of improvising.