sips
grade
reliability
overall
sips is a native macOS command-line tool designed for image querying and transformation. Handing an agent this utility provides a fast, lightweight interface for basic spatial modifications like scaling and cropping. However, because it lacks support for modern compression formats and text compositing, the agent cannot complete web-ready production tasks without integrating external command-line applications into the rendering pipeline.
Surfaces × lenses
| surface ↓ lens → | Coherence | Economy | Consistency | Verifiability | Safety |
|---|---|---|---|---|---|
| Disclosure | |||||
| Interface | |||||
| Loop | |||||
| Recursion | |||||
| Human |
Click any cell for what was measured and why. A corner dot marks a cell filled by a hard instrument (checker, telemetry); undotted cells are judged by the rater ensemble. 53% of this matrix is hard-measured.
The experience
The agent interacts with sips through terminal commands using flags to modify image dimensions and properties. For standard spatial operations like resizing or centered cropping, the tool works well because the commands behave deterministically. Specifying the target output path using the --out option prevents in-place edits on original media, which protects files from accidental corruption or loss during command testing.
However, sips works poorly for modern image delivery pipelines due to severe functional gaps. The utility cannot convert files to WebP formats or apply text watermarks and color tone adjustments. This forces the agent to introduce external dependencies like ImageMagick and cwebp. The resulting pipeline is complex and expensive, requiring multiple separate shell commands to export intermediate files and coordinate transformations across three distinct command-line tools.
Syntactically, the interface contains non-standard conventions, such as specifying target-height before target-width in --cropToHeightWidth, though explicit documentation helps the agent avoid ordering errors. The primary point of friction is the tool's runtime instability; repeated executions of identical pipelines to produce the final asset often fail without clear diagnostic output. This execution unreliability forces agents to engage in redundant verification loops, though the tool adequately supports pipeline resumption because resuming from saved files on disk never corrupts prior output.
Fit
sips is a good fit for agents performing lightweight, localized image manipulation on macOS environments where installing external dependencies is not possible. It serves adequately for quick dimension queries, cropping, and standard JPEG conversions, as its low system overhead and standard UNIX syntax integrate easily with shell execution loops.
It is a poor fit for agents tasked with automated web content generation or high-volume production pipelines. The tool's inability to write modern formats, perform color grading, or apply watermarks requires supplementary utilities, rendering its standalone use insufficient for complete media pipelines.
Notes
This diagram shows how the agent coordinates multiple image processing tasks and handles potential runtime failures when using sips.
Validation
Convergent check. Where a hard instrument and the independent rater panel measure the same cell, they agree on 8 of 16 cross-checked cells (78%). Two independent methods landing on the same number is our accuracy signal — not just the raters agreeing with each other.
| Cell where they disagree | Instrument | Rater | Δ |
|---|---|---|---|
| disclosure.verifiability | -0.40 | +0.60 | 1.00 |
| human.verifiability | -0.20 | +0.80 | 1.00 |
| recursion.coherence | -0.56 | +0.40 | 0.96 |
| disclosure.economy | +0.03 | +0.70 | 0.67 |
| human.safety | -0.10 | +0.50 | 0.60 |
| interface.coherence | -0.11 | +0.40 | 0.51 |
| loop.determinism | +0.10 | +0.60 | 0.50 |
| recursion.economy | +0.72 | +0.30 | 0.42 |
Ground truth
4 of 4 requirements met · deterministic checker, no model judgment- WebP output produced — 71KB
- JPEG fallback produced — 124KB
- cropped to ~1.91:1 frame — 1200×630 (1.90:1)
- each output under 200KB — 71KB, 124KB
Reliability (pass^k): one attempt succeeds 80% of the time · three in a row 40% · all runs 0%. A fresh agent resuming from the files alone never broke prior work.
Models & runs
2026-07-07| Model | Harness | Runs | Solved | Turns |
|---|---|---|---|---|
| | 23 | 20/23 | 7 |
Each path is one run through the shared semantic space; nodes are turns (red = an errored turn). Hover a model — here or in the graph — to isolate its runs.
How this was measured
- reference agent
- pi:z-ai/glm-5.2
- trials
- 25 (n=5/tier)
- methodology
- v0.1 3e74a3c6
- cost
- $1.0323
- hard-measured cells
- 53%
- teaching source
- sips (official docs) (official-docs)
Same brief, same reference agent, and the same trial count for every tool in a category — a difference in the numbers is attributable to the tool (and its teaching source), not the setup. Profiles carry the methodology hash; results under different hashes are not compared.