Menü
How to process images in kiro-cli sessions — kiro-cli vs Claude multimodal limits, sizing, and the fresh-session workaround for context compaction losing inline images.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Basierend auf der SOC-Berufsklassifikation
Dispatch Codex to review all unreviewed repository work, then verify its correlated findings ticket or clean result. Use when requesting an independent Codex review, auditing all changes, checking work since the last review, or waiting for Codex findings. Trigger: dispatch codex review, independent review, review all changes, check codex findings, run review agent.
Always-on: when a project has tickets, work the frontier (first unblocked open ticket), mark done, update plan status.
Incremental and full-history repository review with resumable adoption batches. Use when reviewing new work, initializing review tracking, reviewing all work to date, resuming historical review, inspecting unreviewed commits, or reviewing changed tickets. Trigger: review new work, adopt repository, review all history, resume review, review marker, unreviewed commits.
End-of-session handoff that captures current state for the next session. Use when ending a work session, switching context, or before a long break.
Consolidate project artifacts — promote scratch to memory, deduplicate memory, process decisions, organize scripts, update steering/skills accuracy. Use periodically or when the project feels cluttered.
Run and interpret crew-research evals — run.sh flags, scores.jsonl fields, activation TPR/FPR verdicts, resuming interrupted runs. Use when running evals, reading eval results, diagnosing a failing definition, or resuming a dead run. Trigger: run the evals, eval results, scores.jsonl, activation test, TPR, delta threshold, resume the run, skip-completed, known gap.
| name | image-handling |
| description | How to process images in kiro-cli sessions — kiro-cli vs Claude multimodal limits, sizing, and the fresh-session workaround for context compaction losing inline images. |
| metadata | {"type":"reference","invocation":"agent-only","practice":null,"tool":"kiro-cli"} |
In long sessions or after context compaction, inline image processing may be unavailable. Dispatch a fresh session to analyze images:
kiro-cli chat --no-interactive --trust-tools=read "<specific question> /abs/path/to/image.png"
Validated 2026-07-22 (archwright visual harness birth run): the fresh session reads the path via the read tool's Image mode and returns accurate OCR-level analysis. Headless+image is NOT explicitly documented together by kiro.dev — treat as field-proven, re-validate on major CLI upgrades.
kiro-cli enforces file-level limits; Claude's API enforces pixel/token limits. Respect BOTH — the binding constraint is usually Claude's resize tier.
| Constraint | kiro-cli (kiro.dev/docs/cli/chat/images/) | Claude API (docs.anthropic.com) |
|---|---|---|
| Formats | JPEG/JPG, PNG, GIF, WebP | Same (animated: first frame only) |
| File size | < 10 MB per image | 10 MB base64 (5 MB on Bedrock/Vertex) |
| Images per request | ≤ 10 | 20 (claude.ai) / 100+ (API) — kiro-cli's 10 binds first |
| Pixel dims | not documented | 8000×8000 max; resized to tier below |
| Resize tier | n/a (no auto-resize documented) | standard: 1568 px long edge; high-res: 2576 px (Opus 4.7+/Sonnet 5, automatic) |
| Min useful size | not documented | < 200 px risks hallucination (official guidance) |
Patch-based: ⌈w/28⌉ × ⌈h/28⌉ tokens (28-px patches). Example: 1000×1000 ≈ 1296
tokens. The old w×h/750 estimate is obsolete — do not use it.
Pre-resize to ≤ 1568 px long edge and ≥ 200 px short edge YOURSELF — resolution mismatch with the model's native tier is the #1 documented cause of vision grounding failure, and pre-resizing keeps the crop-vs-scale choice in your hands:
convert image.png -resize '1568x1568>' resized.png # shrink if over tier
For per-tool invocation syntax, dispatch priority, multi-validator consensus, and fallback paths when no vision tool is available, see references/tool-dispatch.md.