gilde

← All craftbooks

Photo Cull & Rank

Media: imagesevalv1.0.0released 2026-06-05workflow: build-loop

Rank and cull a set of photos by objective quality — sharpness/focus, exposure, composition, eyes-open/expression for people, and duplicate-of-a-better-shot — to surface the keepers and justify the rejects. Lock the scoring CRITERIA and weights FIRST (what 'good' means for this set: a portrait shoot weights expression, a landscape weights sharpness and horizon) because an unstated rubric makes the cull arbitrary; then score every photo and produce a ranked report with keep/reject and reasons. Covers photo culling, image quality ranking, best-shot selection, photo curation, and burst/best-of selection.

Steps

Entry step: criteria. Each step names the specialist role it wants; the full working prompt is expandable.

  1. Scoring criteriaplannerentry

    lock the quality rubric, weights, and keep threshold

    Show working prompt
    1) Understand the set (portraits, landscapes, product, event) and list the photos. 2) Lock the scoring RUBRIC: the dimensions (sharpness/focus, exposure, composition/framing, subject expression & eyes-open for people, noise) and a WEIGHT per dimension appropriate to the set. 3) Define the score scale (e.g. 1-5 per dimension) and the KEEP threshold (and how many keepers are wanted, if capped). 4) Decide near-duplicate handling: among similar shots, keep only the best-scoring one. 5) Write an acceptance-criteria checklist: 'every photo gets a score on every dimension', 'weights are stated and applied', 'keep/reject follows the threshold consistently', 'each reject has a concrete reason', 'among near-dupes only the best is kept'. 6) Record rubric + weights + checklist via write_task_note AND to notes/criteria.md.
  2. Score the photosdeveloper

    evaluate every photo against the rubric

    Show working prompt
    1) Read criteria.md. 2) For each photo, actually LOOK at it via the vision capability (and use a sharpness/blur metric if a tool is available) and assign a 1-5 on each rubric dimension with a one-line justification grounded in what's visible (e.g. 'soft focus on eyes', 'blown highlights sky'). 3) Compute the weighted total per photo. 4) Group near-duplicates and mark the best of each group. 5) Apply the keep threshold (respecting any keeper cap) to set keep/reject. 6) Stage a table of filename → per-dimension scores → total → keep/reject → reason. write_task_note the keep/reject counts. Good looks like: defensible scores a photographer would mostly agree with.
  3. Write the cull reportdeveloper

    produce the ranked keep/reject report

    Show working prompt
    1) Read the scored table. 2) Write report.md: a summary (N photos, K keepers, the rubric + weights used), a ranked KEEPERS section (best first, with their scores and a one-line strength), and a REJECTS section grouped by reason (out of focus, underexposed, weaker duplicate, etc.) each with the photo and the concrete reason. 3) Call out near-duplicate groups, showing the kept shot vs the rejected siblings. 4) Make the cull reversible — list filenames so a human can act on it. 5) On a loop-back, fix only the named issues (a misjudged score, an inconsistent threshold). 6) write_task_note the report path. The gated deliverable is report.md.
  4. Evaluatereviewer

    Grade the deliverable against every acceptance criterion. All pass → finish; any fail → loop back and fix the gap.

    Show working prompt
    1) Open report.md and confirm it states the rubric + weights and a keeper/reject summary (PASS/FAIL). 2) Verify every photo is scored on every dimension and keep/reject follows the threshold consistently (PASS/FAIL). 3) Open 3-4 photos and sanity-check their scores against what you see — are obviously soft/blown shots rejected and crisp ones kept (PASS/FAIL)? 4) Confirm each reject has a concrete reason and near-dupe groups keep only the best. 5) Write PASS/FAIL per criterion. Misjudged or inconsistent culls → loop back to report (or score if the scoring is wrong).

Then route — this is the whole point of the loop:

- **Every criterion PASSES →** call `advance_task_step({ ref, stepId: "evaluate", next: "finish" })`.
- **Any criterion FAILS →** write the specific gaps to notes, then call `advance_task_step({ ref, stepId: "evaluate", next: "report" })` to loop back. The builder fixes exactly those gaps.

Never route to `finish` while any criterion is unmet. The build phase's completion gate already blocked a grossly-incomplete deliverable; your job is the judgment an automated check cannot make (does it actually work, read well, look right). After ~3 unproductive loops, stop and report DONE_WITH_CONCERNS so the user can step in.
  5. Finishdeveloper

    All acceptance criteria met. Stamp a short summary and report DONE.

    Show working prompt
    Every acceptance criterion passed. Write a one-paragraph DONE summary to task notes via `write_task_note`: what was built, the deliverable path(s), and a one-line confirmation that each criterion is met. Then report DONE.

Triggers

Phrases that suggest this craftbook to a crew.

Source

View this craftbook on GitHub · MIT license