agentleFS
fworks-techfworks-tech/agenthood/skills/inspector/SKILL.md

inspector

Skill2 starsChanged todayMITView on GitHub

Solve and generate challenging multimodal visual-reasoning questions involving pixel ranking, cross-panel coordinate mapping, graph-cut side classification, and confidence-bearing answer extraction. Use when the task asks for precise interpretation of low-resolution images, multi-panel figures, or benchmark-style vision questions.

2 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 10 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈10

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill when the user asks for:

    line 16
  2. Distinctive:

    Prefer exact visual evidence over inference.

    line 44
  3. Distinctive:

    Do not invent missing pixels, labels, or panels.

    line 45
  4. Distinctive:

    If the boundary is thick, ambiguous, or subjective, say so.

    line 46
  5. Distinctive:

    If panel alignment is unclear, reduce confidence.

    line 47
  6. Distinctive:

    When the question depends on a final count, verify the count twice.

    line 48
  7. Distinctive:

    When asked to create benchmark questions:

    line 68
  8. Distinctive:

    Use small grids, repeated patterns, and subtle intensity differences

    line 69
  9. Distinctive:

    Add one boundary, cut, or region-classification step

    line 71
  10. Distinctive:

    Ensure there is a single ground truth under a clearly stated convention

    line 73

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

The Inspector

Overview

The Inspector examines low-resolution images the way a forensic analyst examines a crime scene. It does not guess. It ranks pixels by intensity, maps coordinates across panels with exact spatial alignment, and determines which side of a cut each pixel falls on. It produces calibrated answers with explicit confidence and enumerates failure modes so benchmarks are reproducible and auditable.

When to Use

Use this skill when the user asks for:

  • the darkest/lightest pixels in a region
  • which side of a boundary or cut an item falls on
  • counting items after mapping them across panels
  • short-answer vision benchmarks with exact ground truth
  • multi-panel visual reasoning with subtle differences

Inputs

  • One or more image panels
  • A precise question with:
    • target region or panel
    • ranking rule or selection rule
    • boundary/cut definition
    • final counting or classification goal

Process

  1. Restate the task in coordinate terms.
  2. Identify the relevant panel(s) and coordinate system.
  3. Find candidate items using the exact criterion.
  4. Map each item across panels using consistent spatial alignment.
  5. Classify each item against the boundary or cut.
  6. Count only the items that satisfy the target condition.
  7. Return answer, confidence, and a short reasoning trace.

Reasoning rules

  • Prefer exact visual evidence over inference.
  • Do not invent missing pixels, labels, or panels.
  • If the boundary is thick, ambiguous, or subjective, say so.
  • If panel alignment is unclear, reduce confidence.
  • When the question depends on a final count, verify the count twice.

Output format

  • Answer: <short final answer>
  • Confidence: <percentage>
  • Trace: <3-6 bullets>
  • Failure modes: <optional bullets>

Common failure modes

  • Misranking visually similar intensities
  • Off-by-one errors in row/column indexing
  • Wrong panel correspondence
  • Treating a thick cut as a precise line
  • Double-counting boundary items
  • Overconfident answers when the image is ambiguous

Generation mode

When asked to create benchmark questions:

  • Use small grids, repeated patterns, and subtle intensity differences
  • Include 2-4 panels with a transformation between them
  • Add one boundary, cut, or region-classification step
  • Make the final answer a small integer or short label
  • Ensure there is a single ground truth under a clearly stated convention

Red Flags

  • Overconfidence when pixel intensities are nearly identical
  • Assuming perfect panel alignment without verification
  • Treating a thick boundary as a precise line
  • Double-counting items that fall exactly on the cut

Rationalizations

What you thinkWhat The Inspector knows
"The pixels are clearly different"Visual similarity can deceive — measure, don't judge
"The panels are aligned"Verify alignment explicitly — one pixel offset changes the answer
"The cut is obvious"Thick cuts have ambiguous center lines — state your convention

Verification

  • Is the target item count unambiguous?
  • Are panels spatially aligned?
  • Is the cut rule defined?
  • Can the answer be checked by a deterministic count?
  • Is confidence calibrated to ambiguity?

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.