playbooks: shared crop/zoom capability for reading dense images

[kyle-sexton]

Several skills read dense screenshots and figures. Build one shared crop/zoom capability instead of per-skill handling; the Sonnet 5.5 chapter and a cross-model playbook note cover the guidance until then.

Follow-on from the Sonnet 5.5 prompting-guide work (plan: docs/topics/sonnet-5-5-prompting-digest/PLAN.md on branch feat/sonnet-5-5-prompting-digest; interview answer Q14).

[kyle-sexton]

This was generated by AI during triage.

Agent Brief

Type: Feature
Summary: Provide one shared crop/zoom capability that skills reading dense screenshots and figures use, instead of each skill handling it alone.

Current behavior:

• Guidance only: the playbooks host skill has a cross-model "Reading dense images" note (grade an unreadable value as unverified, then follow per-model pointers and the upstream crop-tool recipe).
• Per-skill handling exists in places. computer-use:diagnose documents when screen zoom recovers detail. The animation rotoscope review script cuts 1:1 crops sized to a long-edge limit.
• No shared tool or script exists that any skill can call to crop or zoom an image region before reading it.

Desired behavior:
A single capability, owned by one plugin, that takes an image and a region (or a tiling request) and returns a full-resolution crop within the model's image-size limit, and that skills reading dense images call instead of handling it themselves.

Acceptance criteria:

• One capability exists with a documented input (image, region or tile spec) and output (crop files within the documented long-edge limit).
• A test shows that a crop of a known fixture region keeps its source pixels unscaled.
• At least one existing consumer uses it in place of its own handling, or the decision records why each one keeps its own.

Out of scope:

• Changing the reading-dense-images guidance itself.
• Screen-control zoom in computer-use, which re-captures the screen rather than cropping a file.

Routing: human-gated, needs decision

The form and owner are open: a skill, a script, or an MCP tool, and which plugin owns it. Whether the rotoscope crop logic should be generalized or left alone is also open. Proposed work class once those are settled: scoped (C3, feature). It becomes structural if consumers across several plugins are migrated in the same change.
Basis: verified, plugins/playbooks/skills/fable-5/context/reading-dense-images.md (the interim guidance); plugins/computer-use/skills/diagnose/reference/screenshots-and-zoom.md; plugins/animation/skills/rotoscope/scripts/review.py (an existing per-skill crop); no shared crop helper found by grepping crop/zoom/LONG_EDGE across plugins/ md, py, mjs, js and sh files.

Priority: low. The interim guidance the body asks for has already landed (#5767), so nothing is blocked.

Stale reference: the body's PLAN.md pointer on feat/sonnet-5-5-prompting-digest. That branch merged as #5767; the plan and ledger row Q14 are on #5685.